data

.gitignore

@@ -0,0 +1,11 @@
+DUMMY1
+DUMMY2
+DUMMY3
+logs
+__pycache__
+.ipynb_checkpoints
+.*.swp
+
+build
+*.c
+monotonic_align/monotonic_align

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2021 Jaehyeon Kim
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -1,13 +1,58 @@
----
-title: Vits ATR
-emoji: 📚
-colorFrom: pink
-colorTo: pink
-sdk: gradio
-sdk_version: 3.1.4
-app_file: app.py
-pinned: false
-license: afl-3.0
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# VITS: Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech
+
+### Jaehyeon Kim, Jungil Kong, and Juhee Son
+
+In our recent [paper](https://arxiv.org/abs/2106.06103), we propose VITS: Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech.
+
+Several recent end-to-end text-to-speech (TTS) models enabling single-stage training and parallel sampling have been proposed, but their sample quality does not match that of two-stage TTS systems. In this work, we present a parallel end-to-end TTS method that generates more natural sounding audio than current two-stage models. Our method adopts variational inference augmented with normalizing flows and an adversarial training process, which improves the expressive power of generative modeling. We also propose a stochastic duration predictor to synthesize speech with diverse rhythms from input text. With the uncertainty modeling over latent variables and the stochastic duration predictor, our method expresses the natural one-to-many relationship in which a text input can be spoken in multiple ways with different pitches and rhythms. A subjective human evaluation (mean opinion score, or MOS) on the LJ Speech, a single speaker dataset, shows that our method outperforms the best publicly available TTS systems and achieves a MOS comparable to ground truth.
+
+Visit our [demo](https://jaywalnut310.github.io/vits-demo/index.html) for audio samples.
+
+We also provide the [pretrained models](https://drive.google.com/drive/folders/1ksarh-cJf3F5eKJjLVWY0X1j1qsQqiS2?usp=sharing).
+
+** Update note: Thanks to [Rishikesh (ऋषिकेश)](https://github.com/jaywalnut310/vits/issues/1), our interactive TTS demo is now available on [Colab Notebook](https://colab.research.google.com/drive/1CO61pZizDj7en71NQG_aqqKdGaA_SaBf?usp=sharing).
+
+<table style="width:100%">
+  <tr>
+    <th>VITS at training</th>
+    <th>VITS at inference</th>
+  </tr>
+  <tr>
+    <td><img src="resources/fig_1a.png" alt="VITS at training" height="400"></td>
+    <td><img src="resources/fig_1b.png" alt="VITS at inference" height="400"></td>
+  </tr>
+</table>
+
+
+## Pre-requisites
+0. Python >= 3.6
+0. Clone this repository
+0. Install python requirements. Please refer [requirements.txt](requirements.txt)
+    1. You may need to install espeak first: `apt-get install espeak`
+0. Download datasets
+    1. Download and extract the LJ Speech dataset, then rename or create a link to the dataset folder: `ln -s /path/to/LJSpeech-1.1/wavs DUMMY1`
+    1. For mult-speaker setting, download and extract the VCTK dataset, and downsample wav files to 22050 Hz. Then rename or create a link to the dataset folder: `ln -s /path/to/VCTK-Corpus/downsampled_wavs DUMMY2`
+0. Build Monotonic Alignment Search and run preprocessing if you use your own datasets.
+```sh
+# Cython-version Monotonoic Alignment Search
+cd monotonic_align
+python setup.py build_ext --inplace
+
+# Preprocessing (g2p) for your own datasets. Preprocessed phonemes for LJ Speech and VCTK have been already provided.
+# python preprocess.py --text_index 1 --filelists filelists/ljs_audio_text_train_filelist.txt filelists/ljs_audio_text_val_filelist.txt filelists/ljs_audio_text_test_filelist.txt 
+# python preprocess.py --text_index 2 --filelists filelists/vctk_audio_sid_text_train_filelist.txt filelists/vctk_audio_sid_text_val_filelist.txt filelists/vctk_audio_sid_text_test_filelist.txt
+```
+
+
+## Training Exmaple
+```sh
+# LJ Speech
+python train.py -c configs/ljs_base.json -m ljs_base
+
+# VCTK
+python train_ms.py -c configs/vctk_base.json -m vctk_base
+```
+
+
+## Inference Example
+See [inference.ipynb](inference.ipynb)

attentions.py

@@ -0,0 +1,303 @@
+import copy
+import math
+import numpy as np
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+import commons
+import modules
+from modules import LayerNorm
+   
+
+class Encoder(nn.Module):
+  def __init__(self, hidden_channels, filter_channels, n_heads, n_layers, kernel_size=1, p_dropout=0., window_size=4, **kwargs):
+    super().__init__()
+    self.hidden_channels = hidden_channels
+    self.filter_channels = filter_channels
+    self.n_heads = n_heads
+    self.n_layers = n_layers
+    self.kernel_size = kernel_size
+    self.p_dropout = p_dropout
+    self.window_size = window_size
+
+    self.drop = nn.Dropout(p_dropout)
+    self.attn_layers = nn.ModuleList()
+    self.norm_layers_1 = nn.ModuleList()
+    self.ffn_layers = nn.ModuleList()
+    self.norm_layers_2 = nn.ModuleList()
+    for i in range(self.n_layers):
+      self.attn_layers.append(MultiHeadAttention(hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout, window_size=window_size))
+      self.norm_layers_1.append(LayerNorm(hidden_channels))
+      self.ffn_layers.append(FFN(hidden_channels, hidden_channels, filter_channels, kernel_size, p_dropout=p_dropout))
+      self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+  def forward(self, x, x_mask):
+    attn_mask = x_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+    x = x * x_mask
+    for i in range(self.n_layers):
+      y = self.attn_layers[i](x, x, attn_mask)
+      y = self.drop(y)
+      x = self.norm_layers_1[i](x + y)
+
+      y = self.ffn_layers[i](x, x_mask)
+      y = self.drop(y)
+      x = self.norm_layers_2[i](x + y)
+    x = x * x_mask
+    return x
+
+
+class Decoder(nn.Module):
+  def __init__(self, hidden_channels, filter_channels, n_heads, n_layers, kernel_size=1, p_dropout=0., proximal_bias=False, proximal_init=True, **kwargs):
+    super().__init__()
+    self.hidden_channels = hidden_channels
+    self.filter_channels = filter_channels
+    self.n_heads = n_heads
+    self.n_layers = n_layers
+    self.kernel_size = kernel_size
+    self.p_dropout = p_dropout
+    self.proximal_bias = proximal_bias
+    self.proximal_init = proximal_init
+
+    self.drop = nn.Dropout(p_dropout)
+    self.self_attn_layers = nn.ModuleList()
+    self.norm_layers_0 = nn.ModuleList()
+    self.encdec_attn_layers = nn.ModuleList()
+    self.norm_layers_1 = nn.ModuleList()
+    self.ffn_layers = nn.ModuleList()
+    self.norm_layers_2 = nn.ModuleList()
+    for i in range(self.n_layers):
+      self.self_attn_layers.append(MultiHeadAttention(hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout, proximal_bias=proximal_bias, proximal_init=proximal_init))
+      self.norm_layers_0.append(LayerNorm(hidden_channels))
+      self.encdec_attn_layers.append(MultiHeadAttention(hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout))
+      self.norm_layers_1.append(LayerNorm(hidden_channels))
+      self.ffn_layers.append(FFN(hidden_channels, hidden_channels, filter_channels, kernel_size, p_dropout=p_dropout, causal=True))
+      self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+  def forward(self, x, x_mask, h, h_mask):
+    """
+    x: decoder input
+    h: encoder output
+    """
+    self_attn_mask = commons.subsequent_mask(x_mask.size(2)).to(device=x.device, dtype=x.dtype)
+    encdec_attn_mask = h_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+    x = x * x_mask
+    for i in range(self.n_layers):
+      y = self.self_attn_layers[i](x, x, self_attn_mask)
+      y = self.drop(y)
+      x = self.norm_layers_0[i](x + y)
+
+      y = self.encdec_attn_layers[i](x, h, encdec_attn_mask)
+      y = self.drop(y)
+      x = self.norm_layers_1[i](x + y)
+      
+      y = self.ffn_layers[i](x, x_mask)
+      y = self.drop(y)
+      x = self.norm_layers_2[i](x + y)
+    x = x * x_mask
+    return x
+
+
+class MultiHeadAttention(nn.Module):
+  def __init__(self, channels, out_channels, n_heads, p_dropout=0., window_size=None, heads_share=True, block_length=None, proximal_bias=False, proximal_init=False):
+    super().__init__()
+    assert channels % n_heads == 0
+
+    self.channels = channels
+    self.out_channels = out_channels
+    self.n_heads = n_heads
+    self.p_dropout = p_dropout
+    self.window_size = window_size
+    self.heads_share = heads_share
+    self.block_length = block_length
+    self.proximal_bias = proximal_bias
+    self.proximal_init = proximal_init
+    self.attn = None
+
+    self.k_channels = channels // n_heads
+    self.conv_q = nn.Conv1d(channels, channels, 1)
+    self.conv_k = nn.Conv1d(channels, channels, 1)
+    self.conv_v = nn.Conv1d(channels, channels, 1)
+    self.conv_o = nn.Conv1d(channels, out_channels, 1)
+    self.drop = nn.Dropout(p_dropout)
+
+    if window_size is not None:
+      n_heads_rel = 1 if heads_share else n_heads
+      rel_stddev = self.k_channels**-0.5
+      self.emb_rel_k = nn.Parameter(torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels) * rel_stddev)
+      self.emb_rel_v = nn.Parameter(torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels) * rel_stddev)
+
+    nn.init.xavier_uniform_(self.conv_q.weight)
+    nn.init.xavier_uniform_(self.conv_k.weight)
+    nn.init.xavier_uniform_(self.conv_v.weight)
+    if proximal_init:
+      with torch.no_grad():
+        self.conv_k.weight.copy_(self.conv_q.weight)
+        self.conv_k.bias.copy_(self.conv_q.bias)
+      
+  def forward(self, x, c, attn_mask=None):
+    q = self.conv_q(x)
+    k = self.conv_k(c)
+    v = self.conv_v(c)
+    
+    x, self.attn = self.attention(q, k, v, mask=attn_mask)
+
+    x = self.conv_o(x)
+    return x
+
+  def attention(self, query, key, value, mask=None):
+    # reshape [b, d, t] -> [b, n_h, t, d_k]
+    b, d, t_s, t_t = (*key.size(), query.size(2))
+    query = query.view(b, self.n_heads, self.k_channels, t_t).transpose(2, 3)
+    key = key.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+    value = value.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+
+    scores = torch.matmul(query / math.sqrt(self.k_channels), key.transpose(-2, -1))
+    if self.window_size is not None:
+      assert t_s == t_t, "Relative attention is only available for self-attention."
+      key_relative_embeddings = self._get_relative_embeddings(self.emb_rel_k, t_s)
+      rel_logits = self._matmul_with_relative_keys(query /math.sqrt(self.k_channels), key_relative_embeddings)
+      scores_local = self._relative_position_to_absolute_position(rel_logits)
+      scores = scores + scores_local
+    if self.proximal_bias:
+      assert t_s == t_t, "Proximal bias is only available for self-attention."
+      scores = scores + self._attention_bias_proximal(t_s).to(device=scores.device, dtype=scores.dtype)
+    if mask is not None:
+      scores = scores.masked_fill(mask == 0, -1e4)
+      if self.block_length is not None:
+        assert t_s == t_t, "Local attention is only available for self-attention."
+        block_mask = torch.ones_like(scores).triu(-self.block_length).tril(self.block_length)
+        scores = scores.masked_fill(block_mask == 0, -1e4)
+    p_attn = F.softmax(scores, dim=-1) # [b, n_h, t_t, t_s]
+    p_attn = self.drop(p_attn)
+    output = torch.matmul(p_attn, value)
+    if self.window_size is not None:
+      relative_weights = self._absolute_position_to_relative_position(p_attn)
+      value_relative_embeddings = self._get_relative_embeddings(self.emb_rel_v, t_s)
+      output = output + self._matmul_with_relative_values(relative_weights, value_relative_embeddings)
+    output = output.transpose(2, 3).contiguous().view(b, d, t_t) # [b, n_h, t_t, d_k] -> [b, d, t_t]
+    return output, p_attn
+
+  def _matmul_with_relative_values(self, x, y):
+    """
+    x: [b, h, l, m]
+    y: [h or 1, m, d]
+    ret: [b, h, l, d]
+    """
+    ret = torch.matmul(x, y.unsqueeze(0))
+    return ret
+
+  def _matmul_with_relative_keys(self, x, y):
+    """
+    x: [b, h, l, d]
+    y: [h or 1, m, d]
+    ret: [b, h, l, m]
+    """
+    ret = torch.matmul(x, y.unsqueeze(0).transpose(-2, -1))
+    return ret
+
+  def _get_relative_embeddings(self, relative_embeddings, length):
+    max_relative_position = 2 * self.window_size + 1
+    # Pad first before slice to avoid using cond ops.
+    pad_length = max(length - (self.window_size + 1), 0)
+    slice_start_position = max((self.window_size + 1) - length, 0)
+    slice_end_position = slice_start_position + 2 * length - 1
+    if pad_length > 0:
+      padded_relative_embeddings = F.pad(
+          relative_embeddings,
+          commons.convert_pad_shape([[0, 0], [pad_length, pad_length], [0, 0]]))
+    else:
+      padded_relative_embeddings = relative_embeddings
+    used_relative_embeddings = padded_relative_embeddings[:,slice_start_position:slice_end_position]
+    return used_relative_embeddings
+
+  def _relative_position_to_absolute_position(self, x):
+    """
+    x: [b, h, l, 2*l-1]
+    ret: [b, h, l, l]
+    """
+    batch, heads, length, _ = x.size()
+    # Concat columns of pad to shift from relative to absolute indexing.
+    x = F.pad(x, commons.convert_pad_shape([[0,0],[0,0],[0,0],[0,1]]))
+
+    # Concat extra elements so to add up to shape (len+1, 2*len-1).
+    x_flat = x.view([batch, heads, length * 2 * length])
+    x_flat = F.pad(x_flat, commons.convert_pad_shape([[0,0],[0,0],[0,length-1]]))
+
+    # Reshape and slice out the padded elements.
+    x_final = x_flat.view([batch, heads, length+1, 2*length-1])[:, :, :length, length-1:]
+    return x_final
+
+  def _absolute_position_to_relative_position(self, x):
+    """
+    x: [b, h, l, l]
+    ret: [b, h, l, 2*l-1]
+    """
+    batch, heads, length, _ = x.size()
+    # padd along column
+    x = F.pad(x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, length-1]]))
+    x_flat = x.view([batch, heads, length**2 + length*(length -1)])
+    # add 0's in the beginning that will skew the elements after reshape
+    x_flat = F.pad(x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [length, 0]]))
+    x_final = x_flat.view([batch, heads, length, 2*length])[:,:,:,1:]
+    return x_final
+
+  def _attention_bias_proximal(self, length):
+    """Bias for self-attention to encourage attention to close positions.
+    Args:
+      length: an integer scalar.
+    Returns:
+      a Tensor with shape [1, 1, length, length]
+    """
+    r = torch.arange(length, dtype=torch.float32)
+    diff = torch.unsqueeze(r, 0) - torch.unsqueeze(r, 1)
+    return torch.unsqueeze(torch.unsqueeze(-torch.log1p(torch.abs(diff)), 0), 0)
+
+
+class FFN(nn.Module):
+  def __init__(self, in_channels, out_channels, filter_channels, kernel_size, p_dropout=0., activation=None, causal=False):
+    super().__init__()
+    self.in_channels = in_channels
+    self.out_channels = out_channels
+    self.filter_channels = filter_channels
+    self.kernel_size = kernel_size
+    self.p_dropout = p_dropout
+    self.activation = activation
+    self.causal = causal
+
+    if causal:
+      self.padding = self._causal_padding
+    else:
+      self.padding = self._same_padding
+
+    self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size)
+    self.conv_2 = nn.Conv1d(filter_channels, out_channels, kernel_size)
+    self.drop = nn.Dropout(p_dropout)
+
+  def forward(self, x, x_mask):
+    x = self.conv_1(self.padding(x * x_mask))
+    if self.activation == "gelu":
+      x = x * torch.sigmoid(1.702 * x)
+    else:
+      x = torch.relu(x)
+    x = self.drop(x)
+    x = self.conv_2(self.padding(x * x_mask))
+    return x * x_mask
+  
+  def _causal_padding(self, x):
+    if self.kernel_size == 1:
+      return x
+    pad_l = self.kernel_size - 1
+    pad_r = 0
+    padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+    x = F.pad(x, commons.convert_pad_shape(padding))
+    return x
+
+  def _same_padding(self, x):
+    if self.kernel_size == 1:
+      return x
+    pad_l = (self.kernel_size - 1) // 2
+    pad_r = self.kernel_size // 2
+    padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+    x = F.pad(x, commons.convert_pad_shape(padding))
+    return x

commons.py

@@ -0,0 +1,161 @@
+import math
+import numpy as np
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+
+def init_weights(m, mean=0.0, std=0.01):
+  classname = m.__class__.__name__
+  if classname.find("Conv") != -1:
+    m.weight.data.normal_(mean, std)
+
+
+def get_padding(kernel_size, dilation=1):
+  return int((kernel_size*dilation - dilation)/2)
+
+
+def convert_pad_shape(pad_shape):
+  l = pad_shape[::-1]
+  pad_shape = [item for sublist in l for item in sublist]
+  return pad_shape
+
+
+def intersperse(lst, item):
+  result = [item] * (len(lst) * 2 + 1)
+  result[1::2] = lst
+  return result
+
+
+def kl_divergence(m_p, logs_p, m_q, logs_q):
+  """KL(P||Q)"""
+  kl = (logs_q - logs_p) - 0.5
+  kl += 0.5 * (torch.exp(2. * logs_p) + ((m_p - m_q)**2)) * torch.exp(-2. * logs_q)
+  return kl
+
+
+def rand_gumbel(shape):
+  """Sample from the Gumbel distribution, protect from overflows."""
+  uniform_samples = torch.rand(shape) * 0.99998 + 0.00001
+  return -torch.log(-torch.log(uniform_samples))
+
+
+def rand_gumbel_like(x):
+  g = rand_gumbel(x.size()).to(dtype=x.dtype, device=x.device)
+  return g
+
+
+def slice_segments(x, ids_str, segment_size=4):
+  ret = torch.zeros_like(x[:, :, :segment_size])
+  for i in range(x.size(0)):
+    idx_str = ids_str[i]
+    idx_end = idx_str + segment_size
+    ret[i] = x[i, :, idx_str:idx_end]
+  return ret
+
+
+def rand_slice_segments(x, x_lengths=None, segment_size=4):
+  b, d, t = x.size()
+  if x_lengths is None:
+    x_lengths = t
+  ids_str_max = x_lengths - segment_size + 1
+  ids_str = (torch.rand([b]).to(device=x.device) * ids_str_max).to(dtype=torch.long)
+  ret = slice_segments(x, ids_str, segment_size)
+  return ret, ids_str
+
+
+def get_timing_signal_1d(
+    length, channels, min_timescale=1.0, max_timescale=1.0e4):
+  position = torch.arange(length, dtype=torch.float)
+  num_timescales = channels // 2
+  log_timescale_increment = (
+      math.log(float(max_timescale) / float(min_timescale)) /
+      (num_timescales - 1))
+  inv_timescales = min_timescale * torch.exp(
+      torch.arange(num_timescales, dtype=torch.float) * -log_timescale_increment)
+  scaled_time = position.unsqueeze(0) * inv_timescales.unsqueeze(1)
+  signal = torch.cat([torch.sin(scaled_time), torch.cos(scaled_time)], 0)
+  signal = F.pad(signal, [0, 0, 0, channels % 2])
+  signal = signal.view(1, channels, length)
+  return signal
+
+
+def add_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4):
+  b, channels, length = x.size()
+  signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+  return x + signal.to(dtype=x.dtype, device=x.device)
+
+
+def cat_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4, axis=1):
+  b, channels, length = x.size()
+  signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+  return torch.cat([x, signal.to(dtype=x.dtype, device=x.device)], axis)
+
+
+def subsequent_mask(length):
+  mask = torch.tril(torch.ones(length, length)).unsqueeze(0).unsqueeze(0)
+  return mask
+
+
+@torch.jit.script
+def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):
+  n_channels_int = n_channels[0]
+  in_act = input_a + input_b
+  t_act = torch.tanh(in_act[:, :n_channels_int, :])
+  s_act = torch.sigmoid(in_act[:, n_channels_int:, :])
+  acts = t_act * s_act
+  return acts
+
+
+def convert_pad_shape(pad_shape):
+  l = pad_shape[::-1]
+  pad_shape = [item for sublist in l for item in sublist]
+  return pad_shape
+
+
+def shift_1d(x):
+  x = F.pad(x, convert_pad_shape([[0, 0], [0, 0], [1, 0]]))[:, :, :-1]
+  return x
+
+
+def sequence_mask(length, max_length=None):
+  if max_length is None:
+    max_length = length.max()
+  x = torch.arange(max_length, dtype=length.dtype, device=length.device)
+  return x.unsqueeze(0) < length.unsqueeze(1)
+
+
+def generate_path(duration, mask):
+  """
+  duration: [b, 1, t_x]
+  mask: [b, 1, t_y, t_x]
+  """
+  device = duration.device
+  
+  b, _, t_y, t_x = mask.shape
+  cum_duration = torch.cumsum(duration, -1)
+  
+  cum_duration_flat = cum_duration.view(b * t_x)
+  path = sequence_mask(cum_duration_flat, t_y).to(mask.dtype)
+  path = path.view(b, t_x, t_y)
+  path = path - F.pad(path, convert_pad_shape([[0, 0], [1, 0], [0, 0]]))[:, :-1]
+  path = path.unsqueeze(1).transpose(2,3) * mask
+  return path
+
+
+def clip_grad_value_(parameters, clip_value, norm_type=2):
+  if isinstance(parameters, torch.Tensor):
+    parameters = [parameters]
+  parameters = list(filter(lambda p: p.grad is not None, parameters))
+  norm_type = float(norm_type)
+  if clip_value is not None:
+    clip_value = float(clip_value)
+
+  total_norm = 0
+  for p in parameters:
+    param_norm = p.grad.data.norm(norm_type)
+    total_norm += param_norm.item() ** norm_type
+    if clip_value is not None:
+      p.grad.data.clamp_(min=-clip_value, max=clip_value)
+  total_norm = total_norm ** (1. / norm_type)
+  return total_norm

configs/ATR.json

@@ -0,0 +1,54 @@
+{
+    "train": {
+      "log_interval": 100,
+      "eval_interval": 1000,
+      "colab_save_interval":5000,
+      "seed": 1234,
+      "epochs": 20000,
+      "learning_rate": 2e-4,
+      "betas": [0.8, 0.99],
+      "eps": 1e-9,
+      "batch_size": 32,
+      "fp16_run": true,
+      "lr_decay": 0.999875,
+      "segment_size": 8192,
+      "init_lr_ratio": 1,
+      "warmup_epochs": 0,
+      "c_mel": 45,
+      "c_kl": 1.0
+    },
+    "data": {
+      "training_files":"filelists/ATR.txt.cleaned",
+      "validation_files":"filelists/ATR_val.txt.cleaned",
+      "text_cleaners":["japanese_phrase_cleaners"],
+      "max_wav_value": 32768.0,
+      "sampling_rate": 22050,
+      "filter_length": 1024,
+      "hop_length": 256,
+      "win_length": 1024,
+      "n_mel_channels": 80,
+      "mel_fmin": 0.0,
+      "mel_fmax": null,
+      "add_blank": true,
+      "n_speakers": 0,
+      "cleaned_text": true
+    },
+    "model": {
+      "inter_channels": 192,
+      "hidden_channels": 192,
+      "filter_channels": 768,
+      "n_heads": 2,
+      "n_layers": 6,
+      "kernel_size": 3,
+      "p_dropout": 0.1,
+      "resblock": "1",
+      "resblock_kernel_sizes": [3,7,11],
+      "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+      "upsample_rates": [8,8,2,2],
+      "upsample_initial_channel": 512,
+      "upsample_kernel_sizes": [16,16,4,4],
+      "n_layers_q": 3,
+      "use_spectral_norm": false
+    }
+  }
+  

configs/ljs_base.json

@@ -0,0 +1,52 @@
+{
+  "train": {
+    "log_interval": 200,
+    "eval_interval": 1000,
+    "seed": 1234,
+    "epochs": 20000,
+    "learning_rate": 2e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 64,
+    "fp16_run": true,
+    "lr_decay": 0.999875,
+    "segment_size": 8192,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "training_files":"filelists/ljs_audio_text_train_filelist.txt.cleaned",
+    "validation_files":"filelists/ljs_audio_text_val_filelist.txt.cleaned",
+    "text_cleaners":["english_cleaners2"],
+    "max_wav_value": 32768.0,
+    "sampling_rate": 22050,
+    "filter_length": 1024,
+    "hop_length": 256,
+    "win_length": 1024,
+    "n_mel_channels": 80,
+    "mel_fmin": 0.0,
+    "mel_fmax": null,
+    "add_blank": true,
+    "n_speakers": 0,
+    "cleaned_text": true
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0.1,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [8,8,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [16,16,4,4],
+    "n_layers_q": 3,
+    "use_spectral_norm": false
+  }
+}

configs/ljs_nosdp.json

@@ -0,0 +1,53 @@
+{
+  "train": {
+    "log_interval": 200,
+    "eval_interval": 1000,
+    "seed": 1234,
+    "epochs": 20000,
+    "learning_rate": 2e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 64,
+    "fp16_run": true,
+    "lr_decay": 0.999875,
+    "segment_size": 8192,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "training_files":"filelists/ljs_audio_text_train_filelist.txt.cleaned",
+    "validation_files":"filelists/ljs_audio_text_val_filelist.txt.cleaned",
+    "text_cleaners":["english_cleaners2"],
+    "max_wav_value": 32768.0,
+    "sampling_rate": 22050,
+    "filter_length": 1024,
+    "hop_length": 256,
+    "win_length": 1024,
+    "n_mel_channels": 80,
+    "mel_fmin": 0.0,
+    "mel_fmax": null,
+    "add_blank": true,
+    "n_speakers": 0,
+    "cleaned_text": true
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0.1,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [8,8,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [16,16,4,4],
+    "n_layers_q": 3,
+    "use_spectral_norm": false,
+    "use_sdp": false
+  }
+}

configs/nen.json

@@ -0,0 +1,53 @@
+{
+  "train": {
+    "log_interval": 50,
+    "eval_interval": 500,
+    "colab_save_interval":5000,
+    "seed": 1234,
+    "epochs": 20000,
+    "learning_rate": 2e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 32,
+    "fp16_run": true,
+    "lr_decay": 0.999875,
+    "segment_size": 8192,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "training_files":"filelists/nen.txt.cleaned",
+    "validation_files":"filelists/nen_val.txt.cleaned",
+    "text_cleaners":["japanese_phrase_cleaners"],
+    "max_wav_value": 32768.0,
+    "sampling_rate": 22050,
+    "filter_length": 1024,
+    "hop_length": 256,
+    "win_length": 1024,
+    "n_mel_channels": 80,
+    "mel_fmin": 0.0,
+    "mel_fmax": null,
+    "add_blank": true,
+    "n_speakers": 0,
+    "cleaned_text": true
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0.1,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [8,8,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [16,16,4,4],
+    "n_layers_q": 3,
+    "use_spectral_norm": false
+  }
+}

configs/sg.json

@@ -0,0 +1,54 @@
+{
+  "train": {
+    "log_interval": 100,
+    "eval_interval": 1000,
+    "colab_save_interval":5000,
+    "seed": 1234,
+    "epochs": 10000,
+    "learning_rate": 2e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 28,
+    "fp16_run": true,
+    "lr_decay": 0.999875,
+    "segment_size": 8192,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "training_files":"filelists/SG_train.txt.cleaned",
+    "validation_files":"filelists/SG_val.txt.cleaned",
+    "text_cleaners":["japanese_phrase_cleaners"],
+    "max_wav_value": 32768.0,
+    "sampling_rate": 22050,
+    "filter_length": 1024,
+    "hop_length": 256,
+    "win_length": 1024,
+    "n_mel_channels": 80,
+    "mel_fmin": 0.0,
+    "mel_fmax": null,
+    "add_blank": true,
+    "n_speakers": 3,
+    "cleaned_text": true
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0.1,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [8,8,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [16,16,4,4],
+    "n_layers_q": 3,
+    "use_spectral_norm": false,
+    "gin_channels": 256
+  }
+}

configs/vctk_base.json

@@ -0,0 +1,53 @@
+{
+  "train": {
+    "log_interval": 200,
+    "eval_interval": 1000,
+    "seed": 1234,
+    "epochs": 10000,
+    "learning_rate": 2e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 64,
+    "fp16_run": true,
+    "lr_decay": 0.999875,
+    "segment_size": 8192,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "training_files":"filelists/vctk_audio_sid_text_train_filelist.txt.cleaned",
+    "validation_files":"filelists/vctk_audio_sid_text_val_filelist.txt.cleaned",
+    "text_cleaners":["english_cleaners2"],
+    "max_wav_value": 32768.0,
+    "sampling_rate": 22050,
+    "filter_length": 1024,
+    "hop_length": 256,
+    "win_length": 1024,
+    "n_mel_channels": 80,
+    "mel_fmin": 0.0,
+    "mel_fmax": null,
+    "add_blank": true,
+    "n_speakers": 109,
+    "cleaned_text": true
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0.1,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [8,8,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [16,16,4,4],
+    "n_layers_q": 3,
+    "use_spectral_norm": false,
+    "gin_channels": 256
+  }
+}

data_utils.py

@@ -0,0 +1,392 @@
+import time
+import os
+import random
+import numpy as np
+import torch
+import torch.utils.data
+
+import commons 
+from mel_processing import spectrogram_torch
+from utils import load_wav_to_torch, load_filepaths_and_text
+from text import text_to_sequence, cleaned_text_to_sequence
+
+
+class TextAudioLoader(torch.utils.data.Dataset):
+    """
+        1) loads audio, text pairs
+        2) normalizes text and converts them to sequences of integers
+        3) computes spectrograms from audio files.
+    """
+    def __init__(self, audiopaths_and_text, hparams):
+        self.audiopaths_and_text = load_filepaths_and_text(audiopaths_and_text)
+        self.text_cleaners  = hparams.text_cleaners
+        self.max_wav_value  = hparams.max_wav_value
+        self.sampling_rate  = hparams.sampling_rate
+        self.filter_length  = hparams.filter_length 
+        self.hop_length     = hparams.hop_length 
+        self.win_length     = hparams.win_length
+        self.sampling_rate  = hparams.sampling_rate 
+
+        self.cleaned_text = getattr(hparams, "cleaned_text", False)
+
+        self.add_blank = hparams.add_blank
+        self.min_text_len = getattr(hparams, "min_text_len", 1)
+        self.max_text_len = getattr(hparams, "max_text_len", 190)
+
+        random.seed(1234)
+        random.shuffle(self.audiopaths_and_text)
+        self._filter()
+
+
+    def _filter(self):
+        """
+        Filter text & store spec lengths
+        """
+        # Store spectrogram lengths for Bucketing
+        # wav_length ~= file_size / (wav_channels * Bytes per dim) = file_size / (1 * 2)
+        # spec_length = wav_length // hop_length
+
+        audiopaths_and_text_new = []
+        lengths = []
+        for audiopath, text in self.audiopaths_and_text:
+            if self.min_text_len <= len(text) and len(text) <= self.max_text_len:
+                audiopaths_and_text_new.append([audiopath, text])
+                lengths.append(os.path.getsize(audiopath) // (2 * self.hop_length))
+        self.audiopaths_and_text = audiopaths_and_text_new
+        self.lengths = lengths
+
+    def get_audio_text_pair(self, audiopath_and_text):
+        # separate filename and text
+        audiopath, text = audiopath_and_text[0], audiopath_and_text[1]
+        text = self.get_text(text)
+        spec, wav = self.get_audio(audiopath)
+        return (text, spec, wav)
+
+    def get_audio(self, filename):
+        audio, sampling_rate = load_wav_to_torch(filename)
+        if sampling_rate != self.sampling_rate:
+            raise ValueError("{} {} SR doesn't match target {} SR".format(
+                sampling_rate, self.sampling_rate))
+        audio_norm = audio / self.max_wav_value
+        audio_norm = audio_norm.unsqueeze(0)
+        spec_filename = filename.replace(".wav", ".spec.pt")
+        if os.path.exists(spec_filename):
+            spec = torch.load(spec_filename)
+        else:
+            spec = spectrogram_torch(audio_norm, self.filter_length,
+                self.sampling_rate, self.hop_length, self.win_length,
+                center=False)
+            spec = torch.squeeze(spec, 0)
+            torch.save(spec, spec_filename)
+        return spec, audio_norm
+
+    def get_text(self, text):
+        if self.cleaned_text:
+            text_norm = cleaned_text_to_sequence(text)
+        else:
+            text_norm = text_to_sequence(text, self.text_cleaners)
+        if self.add_blank:
+            text_norm = commons.intersperse(text_norm, 0)
+        text_norm = torch.LongTensor(text_norm)
+        return text_norm
+
+    def __getitem__(self, index):
+        return self.get_audio_text_pair(self.audiopaths_and_text[index])
+
+    def __len__(self):
+        return len(self.audiopaths_and_text)
+
+
+class TextAudioCollate():
+    """ Zero-pads model inputs and targets
+    """
+    def __init__(self, return_ids=False):
+        self.return_ids = return_ids
+
+    def __call__(self, batch):
+        """Collate's training batch from normalized text and aduio
+        PARAMS
+        ------
+        batch: [text_normalized, spec_normalized, wav_normalized]
+        """
+        # Right zero-pad all one-hot text sequences to max input length
+        _, ids_sorted_decreasing = torch.sort(
+            torch.LongTensor([x[1].size(1) for x in batch]),
+            dim=0, descending=True)
+
+        max_text_len = max([len(x[0]) for x in batch])
+        max_spec_len = max([x[1].size(1) for x in batch])
+        max_wav_len = max([x[2].size(1) for x in batch])
+
+        text_lengths = torch.LongTensor(len(batch))
+        spec_lengths = torch.LongTensor(len(batch))
+        wav_lengths = torch.LongTensor(len(batch))
+
+        text_padded = torch.LongTensor(len(batch), max_text_len)
+        spec_padded = torch.FloatTensor(len(batch), batch[0][1].size(0), max_spec_len)
+        wav_padded = torch.FloatTensor(len(batch), 1, max_wav_len)
+        text_padded.zero_()
+        spec_padded.zero_()
+        wav_padded.zero_()
+        for i in range(len(ids_sorted_decreasing)):
+            row = batch[ids_sorted_decreasing[i]]
+
+            text = row[0]
+            text_padded[i, :text.size(0)] = text
+            text_lengths[i] = text.size(0)
+
+            spec = row[1]
+            spec_padded[i, :, :spec.size(1)] = spec
+            spec_lengths[i] = spec.size(1)
+
+            wav = row[2]
+            wav_padded[i, :, :wav.size(1)] = wav
+            wav_lengths[i] = wav.size(1)
+
+        if self.return_ids:
+            return text_padded, text_lengths, spec_padded, spec_lengths, wav_padded, wav_lengths, ids_sorted_decreasing
+        return text_padded, text_lengths, spec_padded, spec_lengths, wav_padded, wav_lengths
+
+
+"""Multi speaker version"""
+class TextAudioSpeakerLoader(torch.utils.data.Dataset):
+    """
+        1) loads audio, speaker_id, text pairs
+        2) normalizes text and converts them to sequences of integers
+        3) computes spectrograms from audio files.
+    """
+    def __init__(self, audiopaths_sid_text, hparams):
+        self.audiopaths_sid_text = load_filepaths_and_text(audiopaths_sid_text)
+        self.text_cleaners = hparams.text_cleaners
+        self.max_wav_value = hparams.max_wav_value
+        self.sampling_rate = hparams.sampling_rate
+        self.filter_length  = hparams.filter_length
+        self.hop_length     = hparams.hop_length
+        self.win_length     = hparams.win_length
+        self.sampling_rate  = hparams.sampling_rate
+
+        self.cleaned_text = getattr(hparams, "cleaned_text", False)
+
+        self.add_blank = hparams.add_blank
+        self.min_text_len = getattr(hparams, "min_text_len", 1)
+        self.max_text_len = getattr(hparams, "max_text_len", 190)
+
+        random.seed(1234)
+        random.shuffle(self.audiopaths_sid_text)
+        self._filter()
+
+    def _filter(self):
+        """
+        Filter text & store spec lengths
+        """
+        # Store spectrogram lengths for Bucketing
+        # wav_length ~= file_size / (wav_channels * Bytes per dim) = file_size / (1 * 2)
+        # spec_length = wav_length // hop_length
+
+        audiopaths_sid_text_new = []
+        lengths = []
+        for audiopath, sid, text in self.audiopaths_sid_text:
+            if self.min_text_len <= len(text) and len(text) <= self.max_text_len:
+                audiopaths_sid_text_new.append([audiopath, sid, text])
+                lengths.append(os.path.getsize(audiopath) // (2 * self.hop_length))
+        self.audiopaths_sid_text = audiopaths_sid_text_new
+        self.lengths = lengths
+
+    def get_audio_text_speaker_pair(self, audiopath_sid_text):
+        # separate filename, speaker_id and text
+        audiopath, sid, text = audiopath_sid_text[0], audiopath_sid_text[1], audiopath_sid_text[2]
+        text = self.get_text(text)
+        spec, wav = self.get_audio(audiopath)
+        sid = self.get_sid(sid)
+        return (text, spec, wav, sid)
+
+    def get_audio(self, filename):
+        audio, sampling_rate = load_wav_to_torch(filename)
+        if sampling_rate != self.sampling_rate:
+            raise ValueError("{} {} SR doesn't match target {} SR".format(
+                sampling_rate, self.sampling_rate))
+        audio_norm = audio / self.max_wav_value
+        audio_norm = audio_norm.unsqueeze(0)
+        spec_filename = filename.replace(".wav", ".spec.pt")
+        if os.path.exists(spec_filename):
+            spec = torch.load(spec_filename)
+        else:
+            spec = spectrogram_torch(audio_norm, self.filter_length,
+                self.sampling_rate, self.hop_length, self.win_length,
+                center=False)
+            spec = torch.squeeze(spec, 0)
+            torch.save(spec, spec_filename)
+        return spec, audio_norm
+
+    def get_text(self, text):
+        if self.cleaned_text:
+            text_norm = cleaned_text_to_sequence(text)
+        else:
+            text_norm = text_to_sequence(text, self.text_cleaners)
+        if self.add_blank:
+            text_norm = commons.intersperse(text_norm, 0)
+        text_norm = torch.LongTensor(text_norm)
+        return text_norm
+
+    def get_sid(self, sid):
+        sid = torch.LongTensor([int(sid)])
+        return sid
+
+    def __getitem__(self, index):
+        return self.get_audio_text_speaker_pair(self.audiopaths_sid_text[index])
+
+    def __len__(self):
+        return len(self.audiopaths_sid_text)
+
+
+class TextAudioSpeakerCollate():
+    """ Zero-pads model inputs and targets
+    """
+    def __init__(self, return_ids=False):
+        self.return_ids = return_ids
+
+    def __call__(self, batch):
+        """Collate's training batch from normalized text, audio and speaker identities
+        PARAMS
+        ------
+        batch: [text_normalized, spec_normalized, wav_normalized, sid]
+        """
+        # Right zero-pad all one-hot text sequences to max input length
+        _, ids_sorted_decreasing = torch.sort(
+            torch.LongTensor([x[1].size(1) for x in batch]),
+            dim=0, descending=True)
+
+        max_text_len = max([len(x[0]) for x in batch])
+        max_spec_len = max([x[1].size(1) for x in batch])
+        max_wav_len = max([x[2].size(1) for x in batch])
+
+        text_lengths = torch.LongTensor(len(batch))
+        spec_lengths = torch.LongTensor(len(batch))
+        wav_lengths = torch.LongTensor(len(batch))
+        sid = torch.LongTensor(len(batch))
+
+        text_padded = torch.LongTensor(len(batch), max_text_len)
+        spec_padded = torch.FloatTensor(len(batch), batch[0][1].size(0), max_spec_len)
+        wav_padded = torch.FloatTensor(len(batch), 1, max_wav_len)
+        text_padded.zero_()
+        spec_padded.zero_()
+        wav_padded.zero_()
+        for i in range(len(ids_sorted_decreasing)):
+            row = batch[ids_sorted_decreasing[i]]
+
+            text = row[0]
+            text_padded[i, :text.size(0)] = text
+            text_lengths[i] = text.size(0)
+
+            spec = row[1]
+            spec_padded[i, :, :spec.size(1)] = spec
+            spec_lengths[i] = spec.size(1)
+
+            wav = row[2]
+            wav_padded[i, :, :wav.size(1)] = wav
+            wav_lengths[i] = wav.size(1)
+
+            sid[i] = row[3]
+
+        if self.return_ids:
+            return text_padded, text_lengths, spec_padded, spec_lengths, wav_padded, wav_lengths, sid, ids_sorted_decreasing
+        return text_padded, text_lengths, spec_padded, spec_lengths, wav_padded, wav_lengths, sid
+
+
+class DistributedBucketSampler(torch.utils.data.distributed.DistributedSampler):
+    """
+    Maintain similar input lengths in a batch.
+    Length groups are specified by boundaries.
+    Ex) boundaries = [b1, b2, b3] -> any batch is included either {x | b1 < length(x) <=b2} or {x | b2 < length(x) <= b3}.
+  
+    It removes samples which are not included in the boundaries.
+    Ex) boundaries = [b1, b2, b3] -> any x s.t. length(x) <= b1 or length(x) > b3 are discarded.
+    """
+    def __init__(self, dataset, batch_size, boundaries, num_replicas=None, rank=None, shuffle=True):
+        super().__init__(dataset, num_replicas=num_replicas, rank=rank, shuffle=shuffle)
+        self.lengths = dataset.lengths
+        self.batch_size = batch_size
+        self.boundaries = boundaries
+  
+        self.buckets, self.num_samples_per_bucket = self._create_buckets()
+        self.total_size = sum(self.num_samples_per_bucket)
+        self.num_samples = self.total_size // self.num_replicas
+  
+    def _create_buckets(self):
+        buckets = [[] for _ in range(len(self.boundaries) - 1)]
+        for i in range(len(self.lengths)):
+            length = self.lengths[i]
+            idx_bucket = self._bisect(length)
+            if idx_bucket != -1:
+                buckets[idx_bucket].append(i)
+  
+        for i in range(len(buckets) - 1, 0, -1):
+            if len(buckets[i]) == 0:
+                buckets.pop(i)
+                self.boundaries.pop(i+1)
+  
+        num_samples_per_bucket = []
+        for i in range(len(buckets)):
+            len_bucket = len(buckets[i])
+            total_batch_size = self.num_replicas * self.batch_size
+            rem = (total_batch_size - (len_bucket % total_batch_size)) % total_batch_size
+            num_samples_per_bucket.append(len_bucket + rem)
+        return buckets, num_samples_per_bucket
+  
+    def __iter__(self):
+      # deterministically shuffle based on epoch
+      g = torch.Generator()
+      g.manual_seed(self.epoch)
+  
+      indices = []
+      if self.shuffle:
+          for bucket in self.buckets:
+              indices.append(torch.randperm(len(bucket), generator=g).tolist())
+      else:
+          for bucket in self.buckets:
+              indices.append(list(range(len(bucket))))
+  
+      batches = []
+      for i in range(len(self.buckets)):
+          bucket = self.buckets[i]
+          len_bucket = len(bucket)
+          ids_bucket = indices[i]
+          num_samples_bucket = self.num_samples_per_bucket[i]
+  
+          # add extra samples to make it evenly divisible
+          rem = num_samples_bucket - len_bucket
+          ids_bucket = ids_bucket + ids_bucket * (rem // len_bucket) + ids_bucket[:(rem % len_bucket)]
+  
+          # subsample
+          ids_bucket = ids_bucket[self.rank::self.num_replicas]
+  
+          # batching
+          for j in range(len(ids_bucket) // self.batch_size):
+              batch = [bucket[idx] for idx in ids_bucket[j*self.batch_size:(j+1)*self.batch_size]]
+              batches.append(batch)
+  
+      if self.shuffle:
+          batch_ids = torch.randperm(len(batches), generator=g).tolist()
+          batches = [batches[i] for i in batch_ids]
+      self.batches = batches
+  
+      assert len(self.batches) * self.batch_size == self.num_samples
+      return iter(self.batches)
+  
+    def _bisect(self, x, lo=0, hi=None):
+      if hi is None:
+          hi = len(self.boundaries) - 1
+  
+      if hi > lo:
+          mid = (hi + lo) // 2
+          if self.boundaries[mid] < x and x <= self.boundaries[mid+1]:
+              return mid
+          elif x <= self.boundaries[mid]:
+              return self._bisect(x, lo, mid)
+          else:
+              return self._bisect(x, mid + 1, hi)
+      else:
+          return -1
+
+    def __len__(self):
+        return self.num_samples // self.batch_size

filelists/ATR_val.txt

@@ -0,0 +1,100 @@
+wav/ATR_b101_001.wav|「…………」
+wav/ATR_b101_002.wav|「……」
+wav/ATR_b101_003.wav|「……ッ」
+wav/ATR_b101_004.wav|「……」
+wav/ATR_b101_005.wav|「……？」
+wav/ATR_b101_006.wav|「……」
+wav/ATR_b101_007.wav|（……）
+wav/ATR_b101_008.wav|（ゴボゴボゴボゴボ）
+wav/ATR_b101_009.wav|（――――！！）
+wav/ATR_b101_010.wav|「……」
+wav/ATR_b101_011.wav|「いえ、当然の務めを果たしたまでです」
+wav/ATR_b101_012.wav|「あなた方ヒトがそう総称する精密機器に属していますが、その呼び方は少々広範囲すぎるかと思います」
+wav/ATR_b101_013.wav|「当然です。高性能ですから」
+wav/ATR_b101_014.wav|「そうです、ダメです」
+wav/ATR_b101_015.wav|「わたしはマスターの所有物ですので。勝手に売買するのは違法です」
+wav/ATR_b101_016.wav|「え……」
+wav/ATR_b101_017.wav|「ええぇぇ……マスター死んじゃったんですかぁ。そうですかぁ。合掌」
+wav/ATR_b101_018.wav|「……」
+wav/ATR_b101_019.wav|「……そう……なんですか。あなたが……」
+wav/ATR_b101_020.wav|「さっき、海の中で出逢った時から、そんな気がしていました。やっぱりそうだったんですね」
+wav/ATR_b101_021.wav|「なんとお呼びしましょう。ご希望があればどうぞ」
+wav/ATR_b101_022.wav|「例えば“お兄ちゃん”とか“パパ”とか“にーさま”とか“兄ちゃま”とか……」
+wav/ATR_b101_023.wav|「夏生……」
+wav/ATR_b101_024.wav|「型番ですか？」
+wav/ATR_b101_025.wav|「“アトリ”です」
+wav/ATR_b101_026.wav|「マスターが名付けてくれました」
+wav/ATR_b101_027.wav|「でしょう！固有名が与えられるのは、物であるロボットが命ある者として承認された証ですからね」
+wav/ATR_b101_028.wav|「いわゆるひとつのステータスってやつです♪」
+wav/ATR_b101_029.wav|「あ、“夏生”もなかなかいい名前だと思いますよ」
+wav/ATR_b101_030.wav|「？　こういう目線ですが？」
+wav/ATR_b101_031.wav|「よろしくお願いしますね、夏生さん」
+wav/ATR_b102_001.wav|「……学校」
+wav/ATR_b102_002.wav|「手をお貸ししましょうか？」
+wav/ATR_b102_003.wav|「はいです」
+wav/ATR_b102_004.wav|「質問してもいいですか。ここはどこでしょう」
+wav/ATR_b102_005.wav|「いえ……わたしのメモリーにはありません」
+wav/ATR_b102_006.wav|「へっちゃらです。高性能ですから」
+wav/ATR_b102_007.wav|「どうかしましたか？」
+wav/ATR_b102_008.wav|「？」
+wav/ATR_b102_009.wav|「わたしに買ってくださるんですか？」
+wav/ATR_b102_010.wav|「うーんと……」
+wav/ATR_b102_011.wav|「……待ってください。判断基準が多数あって、決めきれないのです」
+wav/ATR_b102_012.wav|「丈夫さはこれだし、見た目が整ってるのはこれで……お値段とのバランスがよさげなのは……」
+wav/ATR_b102_013.wav|「わぁ～、サイズもぴったりです！」
+wav/ATR_b102_014.wav|「んっふふ～♪」
+wav/ATR_b102_015.wav|「はい！　だって夏生さんがわたしに、物を与えてくださったんですもの！」
+wav/ATR_b102_016.wav|「……嬉しいと変でしたか……？」
+wav/ATR_b102_017.wav|「ですよねー。よかったぁ」
+wav/ATR_b102_018.wav|「わたしは今から売却されるんですよね」
+wav/ATR_b102_019.wav|「あの……お願いがあるんですが」
+wav/ATR_b102_020.wav|「わたしを売るのは少し待っていただけませんか」
+wav/ATR_b102_021.wav|「うんと……４５日です。その後はどうなっても構いません」
+wav/ATR_b102_022.wav|「それは……」
+wav/ATR_b102_023.wav|「……」
+wav/ATR_b102_024.wav|「……」
+wav/ATR_b102_025.wav|「はい、間違いありません。今朝、そちらにいる夏生さんに海から引き揚げてもらいました」
+wav/ATR_b102_026.wav|「寝てたので、覚えてません」
+wav/ATR_b102_027.wav|「それで、わたしは高く売却できそうですか？」
+wav/ATR_b102_028.wav|「なるほど……さすが高性能なわたし。ご満足いただけるプライスが付いたのですね」
+wav/ATR_b102_029.wav|「こちらにいる夏生さんです」
+wav/ATR_b102_030.wav|「？」
+wav/ATR_b102_031.wav|「はい、ありがとうございます」
+wav/ATR_b102_032.wav|「わたしは眠りにつく前、マスターから命令を受けました。それを果たしてからにしてほしいんです」
+wav/ATR_b102_033.wav|「夏生さんは足が必要なんですよね？」
+wav/ATR_b102_034.wav|「それまでの間、わたしが夏生さんの足になります！」
+wav/ATR_b102_035.wav|「こんな立派な靴も買っていただきましたし♪」
+wav/ATR_b102_036.wav|「今使ってるその足より、きっとお役に立ちますよ」
+wav/ATR_b102_037.wav|「どうしてですか？わたしの方が高性能なのに」
+wav/ATR_b102_038.wav|「え、あのっ」
+wav/ATR_b103_001.wav|「寝ながら叫ぶのはヒトとしてなかなか器用な芸ですね……。もしや、脳機能のエラーでしょうか」
+wav/ATR_b103_002.wav|「……夏生さんの脳、修理が必要かも」
+wav/ATR_b103_003.wav|「やってみます」
+wav/ATR_b103_004.wav|「まずはそのびっしょりの汗を拭きますね。風邪をひくと大変です。ヒトは脆くてすぐ故障しますから」
+wav/ATR_b103_005.wav|「泣いてますか？まだ痛いですか？」
+wav/ATR_b103_006.wav|「はい、そうですとも！そこにあるそれなんかより、うんと高性能な」
+wav/ATR_b103_007.wav|「おや……？」
+wav/ATR_b103_008.wav|「あの……夏生さん。汗を拭いてあげたいのですけど」
+wav/ATR_b103_009.wav|「もしかしてわたしで拭いてますか？それは足ではなくタオルの役目ですよ」
+wav/ATR_b103_010.wav|「……。学習しました」
+wav/ATR_b103_011.wav|「いいですよ」
+wav/ATR_b103_012.wav|「ふふ……」
+wav/ATR_b103_013.wav|「おやすみなさい、夏生さん」
+wav/ATR_b103_014.wav|「すやすや…………」
+wav/ATR_b103_015.wav|「いやん……夏生さんのエッチぃ……むふふ♪」
+wav/ATR_b103_016.wav|「すやすや……にへへ……夏生さぁん……」
+wav/ATR_b102_080_kai.wav|「夏生さんは、足を取り戻したら、何がしたかったんですか？」
+wav/ATR_b103_019.wav|「うぅ～ん…………むにゃむにゃ…………夏生さんのアホぉ」
+wav/ATR_b103_020.wav|「うへへ……くすぐったいですぅ」
+wav/ATR_b103_021.wav|「アトリですぅ……それロボットへの[べっしょう,1]蔑称ですからぁ……警告２回目ですよぉ……お仕置きのロケットパン……ツ……」
+wav/ATR_b103_022.wav|「はわっ……！？」
+wav/ATR_b103_023.wav|「あ、そっか……わたし……昨日の夜、ここで夏生さんに抱かれたんでした……」
+wav/ATR_b103_024.wav|「あ……おはようございますぅ、夏生さん……ふわぁ～」
+wav/ATR_b103_025.wav|「わたし低電圧なんでぇ……むにゃむにゃ」
+wav/ATR_b103_026.wav|「ぅぅ、変な姿勢でスリープしてたから、関節がグキグキします……」
+wav/ATR_b103_027.wav|「ふわぁい」
+wav/ATR_b103_028.wav|「？」
+wav/ATR_b103_029.wav|「でしょう！高性能ですからムフン！」
+wav/ATR_b111_001.wav|「シャカシャカシャコシャコ」
+wav/ATR_b111_002.wav|「シャカシャカシャコシャコ……オエッ」
+wav/ATR_b111_003.wav|「はふひはん（夏生さん）」

filelists/ATR_val.txt.cleaned

@@ -0,0 +1,100 @@
+wav/ATR_b101_001.wav|............
+wav/ATR_b101_002.wav|......
+wav/ATR_b101_003.wav|...... cl
+wav/ATR_b101_004.wav|......
+wav/ATR_b101_005.wav|......?
+wav/ATR_b101_006.wav|......
+wav/ATR_b101_007.wav|(......)
+wav/ATR_b101_008.wav|( gobogobo gobogobo)
+wav/ATR_b101_009.wav|(--------!!)
+wav/ATR_b101_010.wav|......
+wav/ATR_b101_011.wav|ie, toozeNno tsUtomeo hatashItamadedesU
+wav/ATR_b101_012.wav|anatakata hItoga soo sooshoo suru seemitsU kikini zokushIte imasUga, sono yobikatawa shooshoo koohaNi sugirukato omoimasU
+wav/ATR_b101_013.wav|toozeNdesU. kooseenoodesUkara
+wav/ATR_b101_014.wav|soodesU, damedesU
+wav/ATR_b101_015.wav|watashiwa masUtaano shoyuumonodesUnode. kaclteni baibai suru nowa ihoodesU
+wav/ATR_b101_016.wav|e......
+wav/ATR_b101_017.wav|eeee...... masUtaa shiNjaclta NdesUkaa. soodesUkaa. gaclshoo
+wav/ATR_b101_018.wav|......
+wav/ATR_b101_019.wav|...... soo...... naNdesUka. anataga......
+wav/ATR_b101_020.wav|saclki, umino nakade deaclta tokikara, soNna kiga shIte imashIta. yaclpari soodaclta NdesUne
+wav/ATR_b101_021.wav|naNto oyobi shimashoo. gokibooga areba doozo
+wav/ATR_b101_022.wav|tatoeba" oniichaN" toka" papa" toka" niisama" toka" anicha m a" toka......
+wav/ATR_b101_023.wav|natsuo......
+wav/ATR_b101_024.wav|katabaNdesUka?
+wav/ATR_b101_025.wav|" atori" desU
+wav/ATR_b101_026.wav|masUtaaga nazukete kuremashIta
+wav/ATR_b101_027.wav|deshoo! koyuumeega ataerareru nowa, monodearu robocltoga inochi aru monoto shIte shooniN sareta akashidesUkarane
+wav/ATR_b101_028.wav|iwayuru hItotsuno sUteetasuclte yatsudesU
+wav/ATR_b101_029.wav|a," natsuo" m o nakanaka ii namaedato omoimasUyo
+wav/ATR_b101_030.wav|? kooyuu meseNdesUga?
+wav/ATR_b101_031.wav|yoroshIkuonegaishimasUne, natsuosaN
+wav/ATR_b102_001.wav|...... gaclkoo
+wav/ATR_b102_002.wav|teo okashI shimashooka?
+wav/ATR_b102_003.wav|haidesU
+wav/ATR_b102_004.wav|shItsumoN shItemo iidesUka. kokowa dokodeshoo
+wav/ATR_b102_005.wav|ie...... watashino memoriiniwaarimaseN
+wav/ATR_b102_006.wav|heclcharadesU. kooseenoodesUkara
+wav/ATR_b102_007.wav|dooka shimashItaka?
+wav/ATR_b102_008.wav|?
+wav/ATR_b102_009.wav|watashini kaclte kudasaru NdesUka?
+wav/ATR_b102_010.wav|uuNto......
+wav/ATR_b102_011.wav|...... maclte kudasai. haNdaNkijuNga tasuu aclte, kimekirenai nodesU
+wav/ATR_b102_012.wav|joobusawa koredashi, mitamega totonoclteru nowa korede...... onedaNtono baraNsuga yosagena nowa......
+wav/ATR_b102_013.wav|waa~, saizumo picltaridesU!
+wav/ATR_b102_014.wav|Nclfufu~
+wav/ATR_b102_015.wav|hai! daclte natsuosaNga watashini, monoo ataete kudasaclta NdesU mono!
+wav/ATR_b102_016.wav|...... ureshiito heNdeshItaka......?
+wav/ATR_b102_017.wav|desUyonee. yokaclta a
+wav/ATR_b102_018.wav|watashiwa imakara baikyakU sareru NdesUyone
+wav/ATR_b102_019.wav|ano...... onegaiga aru NdesUga
+wav/ATR_b102_020.wav|watashio uru nowa sUkoshi maclte itadakemaseNka
+wav/ATR_b102_021.wav|uNto...... yoNjuu gonichidesU. sonogowa doo nacltemo kamaimaseN
+wav/ATR_b102_022.wav|sorewa......
+wav/ATR_b102_023.wav|......
+wav/ATR_b102_024.wav|......
+wav/ATR_b102_025.wav|hai, machigai arimaseN. kesa, sochirani iru natsuosaNni umikara hIkiagete moraimashIta
+wav/ATR_b102_026.wav|netetanode, oboetemaseN
+wav/ATR_b102_027.wav|sorede, watashiwa takaku baikyaku dekisoodesUka?
+wav/ATR_b102_028.wav|naruhodo...... sasuga kooseenoona watashi. gomaNzoku itadakeru puraisuga tsuita nodesUne
+wav/ATR_b102_029.wav|kochirani iru natsuosaNdesU
+wav/ATR_b102_030.wav|?
+wav/ATR_b102_031.wav|hai, arigatoo gozaimasU
+wav/ATR_b102_032.wav|watashiwa nemurini tsuku mae, masUtaakara meereeo ukemashIta. soreo hatashItekarani shItehoshii NdesU
+wav/ATR_b102_033.wav|natsuosaNwa ashiga hItsuyoona NdesUyone?
+wav/ATR_b102_034.wav|soremadeno aida, watashiga natsuosaNno ashini narimasU!
+wav/ATR_b102_035.wav|koNna riclpana kUtsumo kaclte itadakimashItashi
+wav/ATR_b102_036.wav|ima tsUkaclteru sono ashiyori, kiclto oyakunitachimasUyo
+wav/ATR_b102_037.wav|dooshItedesUka? watashino hooga kooseenoonanoni
+wav/ATR_b102_038.wav|e, anocl
+wav/ATR_b103_001.wav|nenagara sakebu nowa hItoto shIte nakanaka kiyoona geedesUne....... moshiya, nookinoono eraadeshooka
+wav/ATR_b103_002.wav|...... natsuosaNno noo, shuuriga hItsuyookamo
+wav/ATR_b103_003.wav|yaclte mimasU
+wav/ATR_b103_004.wav|mazuwa sono biclshorino aseo fUkimasUne. kazeo hIkuto taiheNdesU. hItowa morokUte sugu koshoo shimasUkara
+wav/ATR_b103_005.wav|naitemasUka? mada itaidesUka?
+wav/ATR_b103_006.wav|hai, soodesUtomo! sokoni aru sorenaNkayori, uNtokooseenoona
+wav/ATR_b103_007.wav|oya......?
+wav/ATR_b103_008.wav|ano...... natsuosaN. aseo fuite agetai nodesUkedo
+wav/ATR_b103_009.wav|moshIkashIte watashide fuitemasUka? sorewa ashidewanakU taoruno yakumedesUyo
+wav/ATR_b103_010.wav|....... gakUshuu shimashIta
+wav/ATR_b103_011.wav|iidesUyo
+wav/ATR_b103_012.wav|fufu......
+wav/ATR_b103_013.wav|oyasuminasai, natsuosaN
+wav/ATR_b103_014.wav|suyasuya............
+wav/ATR_b103_015.wav|iyaN...... natsuosaNno eclchii...... mufufu
+wav/ATR_b103_016.wav|suyasuya...... n i hehe...... natsuosaaN......
+wav/ATR_b102_080_kai.wav|natsuosaNwa, ashio torimodoshItara, naniga shItakaclta NdesUka?
+wav/ATR_b103_019.wav|uu~ N............ munya munya............ natsuosaNno ahoo
+wav/ATR_b103_020.wav|uhehe...... kUsugucltaide suu
+wav/ATR_b103_021.wav|atoride suu...... sorerobocltoeno[ beclshoo, ichi] beclshoodesUkara a...... keekoku nikaimedesUyo o...... oshiokino rokecltopaN...... ts u......
+wav/ATR_b103_022.wav|w a wacl......!?
+wav/ATR_b103_023.wav|a, soclka...... watashi...... kinoono yoru, kokode natsuosaNni idakareta NdeshIta......
+wav/ATR_b103_024.wav|a...... ohayoogozaimasU u, natsuosaN...... fuwaa~
+wav/ATR_b103_025.wav|watashI teedeNatsunaNde e...... munya munya
+wav/ATR_b103_026.wav|uu, heNna shIseede suriipu shItetakara, kaNsetsuga gukigukI shimasU......
+wav/ATR_b103_027.wav|fuwaa i
+wav/ATR_b103_028.wav|?
+wav/ATR_b103_029.wav|deshoo! kooseenoodesUkara mufuN!
+wav/ATR_b111_001.wav|shakashaka shakoshako
+wav/ATR_b111_002.wav|shakashaka shakoshako...... oecl
+wav/ATR_b111_003.wav|w a fuhihaN( natsuosaN)

filelists/SG_val.txt

@@ -0,0 +1,300 @@
+SG/MAY_0000.wav|0|ねぇねぇ。なにブツブツ言ってるのー?
+SG/MAY_0001.wav|0|オカリン? ねぇってばー
+SG/MAY_0002.wav|0|誰かと電話中?
+SG/OKA_0027.wav|2|⋯⋯いや、こちらの話だ。問題ない、これより会場に潜入する
+SG/OKA_0028.wav|2|ああ、ドクター中鉢なかばちは抜け駆けをした。たっぷりとその考えについて聞かせてもらうつもりさ
+SG/OKA_0029.wav|2|⋯⋯なに!? 機関が動き出しているだと!?
+SG/OKA_0030.wav|2|そうか、それが運命石の扉シュタインズゲートの選択か。エル・プサイ・コングルゥ
+SG/MAY_0003.wav|0|さっき、ケータイで誰と話してたのー?
+SG/OKA_0031.wav|2|聞くな。それがまゆりのためでもある
+SG/MAY_0004.wav|0|そうなんだー。オカリン、ありがとー
+SG/MAY_0005.wav|0|それよりオカリンオカリン
+SG/OKA_0032.wav|2|まゆりよ、いつも言っているだろう。俺のことをオカリンと呼ぶなと
+SG/MAY_0006.wav|0|えー? でも昔からそう呼んでたよ?
+SG/OKA_0033.wav|2|それは昔の話だ。今の俺は鳳凰院ほうおういん 凶真きょうま。世界中の秘密組織から狙われる、狂気のマッドサイエンティストだ。フゥーハハハ!
+SG/MAY_0007.wav|0|だって、難しくて覚えられないし
+SG/MAY_0008.wav|0|それに、岡部おかべ倫太郎りんたろうと1文字も合ってないよー? おかしいね、えっへへー
+SG/MAY_0009.wav|0|でね、オカリン。えっと、教えてほしいんだけど
+SG/MAY_0010.wav|0|これからここで、なにが始まるのー?
+SG/OKA_0034.wav|2|お前は、それも知らずここまで俺についてきたというのか
+SG/MAY_0011.wav|0|うん
+SG/OKA_0035.wav|2|これからここで始まるのは、ドクター中鉢の記者会見だ
+SG/MAY_0012.wav|0|記者会見? でもー、記者さんなんて見当たらない気がするよ?
+SG/OKA_0036.wav|2|あるいは、機関によるなんらかの妨害を受けたのかもしれないな
+SG/OKA_0037.wav|2|卷き込まれるのは、勘弁だがな
+SG/MAY_0013.wav|0|卷き卷きトカゲ? あ、それを言うならエリ卷きトカゲだねー。えっへへー
+SG/OKA_0038.wav|2|まゆり、気を付けろ。おそらくこの記者会見、なんらかの事件が起き––
+SG/MAY_0014.wav|0|地震かなぁ? 震度2? マグニチュード2? 震度とマグニチュードってどう違うんだっけー⋯⋯
+SG/OKA_0039.wav|2|爆発⋯⋯だと!?
+SG/OKA_0040.wav|2|なんだ⋯⋯これは?
+SG/OKA_0041.wav|2|これは匂う。陰謀の匂いだ。なにを隠したいんだ? さっきの爆発はなんだ?
+SG/OKA_0042.wav|2|俺だ。どうもイヤな予感がする。俺たちが知らないところでなにかが起こっているようだ
+SG/OKA_0043.wav|2|⋯⋯ああ、分かってる。無茶はしないさ。俺も命は惜しいからな。エル・プサイ・コングルゥ
+SG/OKA_0044.wav|2|まゆり、なにをしている?
+SG/MAY_0015.wav|0|んー?
+SG/MAY_0016.wav|0|あのね、うーぱがほしいなあって
+SG/OKA_0045.wav|2|やればいい。うーぱが当たるかどうかは保証できないがな
+SG/MAY_0017.wav|0|でもね、まゆしぃは今、100円玉を切らしちゃっているのです
+SG/MAY_0018.wav|0|だから、オカリンオカリン、100円貸してー?
+SG/OKA_0046.wav|2|甘ったれるなまゆり。金は貸さん。俺がお前に人生の厳しさを教えてやる
+SG/MAY_0019.wav|0|あ、ああー⋯⋯
+SG/MAY_0020.wav|0|あっ、うーぱだよ。しかもメタル。メタルうーぱ
+SG/OKA_0047.wav|2|それはレアなのか?
+SG/MAY_0021.wav|0|すごく!
+SG/OKA_0048.wav|2|フン、ではまゆりにくれてやろう
+SG/MAY_0022.wav|0|ホントー? いいの? オカリン
+SG/OKA_0049.wav|2|鳳凰院凶真だ
+SG/MAY_0023.wav|0|えっへへー、ありがとーオカリン♪
+SG/OKA_0050.wav|2|⋯⋯⋯⋯
+SG/OKA_0051.wav|2|どうやら始まるようだな
+SG/OKA_0052.wav|2|行くぞまゆり
+SG/MAY_0024.wav|0|んー、待って待って。名前書かなくちゃ
+SG/MAY_0025.wav|0|タイムマシン? あの人が作ったのー?
+SG/OKA_0053.wav|2|ドぉぉぉクぅぅぅターぁぁぁっ!
+SG/OKA_0054.wav|2|バカにするにもほどがあるぞ!
+SG/OKA_0055.wav|2|俺が誰なのかはどうでもいい! それより、今貴方が語ったタイムマシンの理論はいったいなんだ!?
+SG/OKA_0056.wav|2|ジョン・タイターのパクリではないか! 貴方はそれでも発明家かっ!
+SG/OKA_0057.wav|2|出ていくのは貴方だ、ドクター! 恥を知れ! 金輪際こんりんざい、貴方には発明家を名乗る資格はないぞっ!
+SG/OKA_0058.wav|2|は、な、せっ⋯⋯んん?
+SG/OKA_0059.wav|2|あ⋯⋯
+SG/OKA_0060.wav|2|き、貴様、機関の人間か!?
+SG/OKA_0061.wav|2|くっ、まさかここまで手が回っているとは⋯⋯。俺としたことが
+SG/OKA_0062.wav|2|⋯⋯⋯⋯
+SG/OKA_0063.wav|2|ここで俺��なにかすれば人目に付くぞ。そうなれば貴様も色々まずいだろう
+SG/OKA_0064.wav|2|それに答える義理はない。機関のやり方は分かっている
+SG/OKA_0065.wav|2|俺だ。機関のエージェントに捕まった。⋯⋯ああ、牧瀬紅莉栖だ、あの女には気を付けろ⋯⋯いや問題ない、ここはなんとか切り抜け––
+SG/OKA_0066.wav|2|くっ、なにをする!
+SG/OKA_0067.wav|2|⋯⋯⋯⋯
+SG/OKA_0068.wav|2|き、貴様に答える義理はないが一応教えてやろう。それは俺以外が触ると自動的に電源がオフになる、特別製のケータイなのだっ。フゥーハハハ!
+SG/OKA_0069.wav|2|⋯⋯っ
+SG/OKA_0070.wav|2|さっきとはいつのことだ?
+SG/OKA_0071.wav|2|俺はすべてお見通しなのだ
+SG/OKA_0072.wav|2|天才少女よ、次会うときは敵同士だな!
+SG/OKA_0073.wav|2|さらばだ、フゥーハハハ!
+SG/OKA_0074.wav|2|き、機関め、あれほどのエージェントを送り込んでくるとは、ついに本気になったようだな⋯⋯!
+SG/OKA_0075.wav|2|だ、だが、俺はまだヤツらに捕まるわけにはいかんのだ⋯⋯
+SG/OKA_0076.wav|2|チィッ! まゆりを置いてきた⋯⋯!
+SG/OKA_0077.wav|2|ん⋯⋯? メールか?
+SG/OKA_0078.wav|2|⋯⋯?
+SG/OKA_0079.wav|2|くっ、まゆり。なぜ出ない
+SG/OKA_0080.wav|2|ハッ、まさか牧瀬紅莉栖め、まゆりをさらったな⋯⋯!
+SG/OKA_0081.wav|2|おのれぇぇ。それが機関のやり方か⋯⋯!
+SG/OKA_0082.wav|2|戻るしかないか⋯⋯
+SG/OKA_0083.wav|2|くくく、あの女、俺に怖じ気づいたか
+SG/OKA_0084.wav|2|よかろう。今曰のところは見逃してやるとしよう
+SG/OKA_0085.wav|2|まゆり、なぜ電話に出ない。そろそろ帰るぞ
+SG/MAY_0026.wav|0|あ、オカリン。メタルうーぱがいなくなっちゃった
+SG/OKA_0086.wav|2|いなくなった? 勝手に動き出したのか。それは実にファンタジーだな
+SG/MAY_0027.wav|0|落としたみたい⋯⋯
+SG/OKA_0087.wav|2|見つからないなら諦めろ。また当てればいい
+SG/MAY_0028.wav|0|当たりっこないよ。だってね、メタルうーぱはネットオークションで、1万円近いプレミアが付いてるんだよ?
+SG/OKA_0088.wav|2|な⋯⋯に⋯⋯?
+SG/OKA_0089.wav|2|まゆりよ、いったいどこで落としたのだ!?
+SG/MAY_0029.wav|0|分かんないから探してるんだよぅ⋯⋯。あと、見つけても、売らないからねー?
+SG/OKA_0090.wav|2|フハハ、その1万円、この俺の研究資金にしてやる
+SG/MAY_0030.wav|0|だから、売らないってばー。まゆしぃの名前書いちゃったし
+SG/MAY_0031.wav|0|トゥットゥルー♪ うーぱさんうーぱさん、出ておいでー
+SG/OKA_0091.wav|2|おのれ、金にしか興味のない下劣なヤツめ、恥を知れ⋯⋯!
+SG/MAY_0032.wav|0|オカリンもねー
+SG/OKA_0093.wav|2|!?
+SG/MAY_0033.wav|0|悲鳴⋯⋯かな?
+SG/OKA_0094.wav|2|まゆり、ここにいろ
+SG/OKA_0095.wav|2|ひっ⋯⋯
+SG/OKA_0096.wav|2|え、な、なんで⋯⋯?
+SG/MAY_0034.wav|0|オカリン、どうしたの⋯⋯?
+SG/OKA_0097.wav|2|で、出るぞっ
+SG/OKA_0098.wav|2|はあ、はあ⋯⋯
+SG/MAY_0035.wav|0|ねぇねぇ、なにがあったのー? 顔色、すごく悪いけど⋯⋯
+SG/OKA_0099.wav|2|人が⋯⋯死んでた
+SG/MAY_0036.wav|0|え⋯⋯
+SG/OKA_0100.wav|2|––っ
+SG/OKA_0101.wav|2|消えた⋯⋯
+SG/OKA_0102.wav|2|おい、そこの貴様。俺たちが見えているか?
+SG/OKA_0103.wav|2|⋯⋯なぜなにも答えない。貴様に聞いているんだぞ? モニタのそっち側にいる、貴様にだ
+SG/OKA_0104.wav|2|ふん。間抜け面をしおって。つまらんヤツだ
+SG/OKA_0105.wav|2|貴様からだと、俺たちはテレビのモニタの中にいるように見えるだろうな。ククク、だがそれは大きな間違いだ
+SG/OKA_0106.wav|2|モニタの中にいるのは貴様なのだよ。貴様が現実だと思っているその世界は、実はすべて虚構。もちろん貴様自身もな
+SG/OKA_0107.wav|2|真の現実、それはこちら側にある
+SG/OKA_0108.wav|2|自分がなにを指摘されているのかすら分かっていないか。無理もない
+SG/OKA_0109.wav|2|まあいい。貴様には分かりやすく、俺たちのことについて説明してやろうではないか
+SG/OKA_0110.wav|2|まず、俺たちがいるのはどこかということだ。ここは東京、秋葉原にある、未来ガジェット研究所だ。俺たちは普段ラボと呼んでいる。世界の支配構造を作り替えるという我が野望の拠点だな
+SG/MAY_0037.wav|0|そうなんだー。悪いことしちゃダメなんだよ、オカリン
+SG/OKA_0111.wav|2|まゆりは少し黙っていろ
+SG/OKA_0112.wav|2|駅から出たら中央通りを進み、末広町駅の交差点を蔵前橋通りへ左折。次の信号の一歩手前の路地を入ると、大檜山���ルという古くさい雑居ビルがある。その2階に我がラボは居を構えている
+SG/OKA_0113.wav|2|目印は、ビル1階にあるブラウン管工房というマニアックなテナントだな
+SG/OKA_0114.wav|2|今どき、旧式のブラウン管テレビだけを扱っているショップだ。いかに電気街である秋葉原と言えども需要があるとは思えない、寂れた店なのだ
+SG/OKA_0115.wav|2|だがブラウン管工房店長である天王寺てんのうじというおっさんは、このビルのオーナーでもある
+SG/OKA_0116.wav|2|故に、今や急ピッチで都市開発が進み、地価も高騰してるこの秋葉原であろうと、道楽丸出しの店を構えていられるというわけだ
+SG/OKA_0117.wav|2|幸いにもあの店長は、人を見る目はあるようでな。この俺のカリスマ性を見抜き、ビル2階をまるまるワンフロア、タダ同然で貸してくれたのだ。フゥーハハハ!
+SG/OKA_0118.wav|2|だが未来ガジェット研究所は深刻な人材不足のため、優秀な研究員を随時募集中だ。今のところ、所属研究員は––
+SG/MAY_0038.wav|0|オカリンオカリン、そこはラボメンって言わなきゃー。所属研究員じゃなくて
+SG/OKA_0119.wav|2|⋯⋯ラボメン、すなわちラボラトリーメンバーは、俺を含めて3人である
+SG/OKA_0120.wav|2|ラボメンナンバー001、ラボ創設者にして狂気のマッドサイエンティストであるこの俺、鳳凰院凶真
+SG/MAY_0039.wav|0|オカリンって呼び方の方がかわいいのにー
+SG/OKA_0121.wav|2|そしてコスプレが趣味の紅一点、ラボメンナンバー002、椎名まゆり
+SG/MAY_0040.wav|0|トゥットゥルー♪ まゆしぃでーす。着るんじゃなくて作るのが趣味だよ
+SG/OKA_0122.wav|2|最後にスーパーハカー、ラボメンナンバー003、橋田はしだ 至いたるだ
+SG/OKA_0123.wav|2|そんな我々3人で構成される未来ガジェット研究所の活動内容は、そのものずばり発明である
+SG/OKA_0124.wav|2|詳細は我がラボのホームページを見てくれ
+SG/OKA_0125.wav|2|もちろん闇の支配権力と戦うための未来ガジェットが最優先事項だが、その研究から派生する副産物的な発明も多い。というか今のところはそっちばかりだ
+SG/OKA_0126.wav|2|すでに我々は8つの未来ガジェットを完成させた。だがこれはまだ序章でしかない。未来ガジェットのアイデアは、俺の中に108まであるのだ
+SG/OKA_0127.wav|2|人の煩悩の数と言え、この@ちゃんねる脳め
+SG/OKA_0128.wav|2|それと、俺が話しているのだから口出しをするなと言っているだろう
+SG/OKA_0129.wav|2|独り言ではない。見て分からないのか。俺は今、モニタの向こうにいるこいつに話しかけているのだ
+SG/MAY_0041.wav|0|あ、今その人、ニヤリって笑ったー
+SG/OKA_0130.wav|2|おのれ貴様、なにを笑っているのか! モニタの中の存在のくせに!
+SG/MAY_0042.wav|0|通じないんじゃないかなー?
+SG/OKA_0131.wav|2|俺たちに話しかけられていることにすら、気付いていないらしいな。自覚がないというのは、実に不幸なことだ
+SG/MAY_0043.wav|0|その人にしてみたら、まゆしぃたちがゲームみたいに見えてるのかなー?
+SG/MAY_0044.wav|0|んじゃ、ダルくんの大好きな2次元の女の子たちもそうなのー?
+SG/OKA_0132.wav|2|ダルの嫁の話はどうでもいい
+SG/OKA_0133.wav|2|ないな
+SG/OKA_0134.wav|2|故に、そのような議論は不毛。世界の支配構造を打ち砕く方法について考える方が、よほど有意義だ
+SG/OKA_0135.wav|2|黙れスーパーハカー。俺は厨二病ではない
+SG/OKA_0136.wav|2|鳳凰院⋯⋯凶真だっ!
+SG/OKA_0137.wav|2|やれやれ。ダルの、人とのコミュニケーションの取れなさは、どれだけ経っても治らないな
+SG/MAY_0045.wav|0|あう⋯⋯。針、指に刺さった⋯⋯
+SG/MAY_0046.wav|0|まゆしぃはオカリンの人質だから、ここにいようと思いまーす
+SG/OKA_0138.wav|2|いや、全然
+SG/OKA_0139.wav|2|この無愛想アルパカめが
+SG/OKA_0140.wav|2|ん?
+SG/MAY_0047.wav|0|きっとね、アルパカさんが、怒っちゃったんだよー
+SG/OKA_0141.wav|2|くっ。後で修理できるか聞きに行かなければ
+SG/OKA_0142.wav|2|消えた⋯⋯
+SG/MAY_0048.wav|0|どうかしたー?
+SG/OKA_0143.wav|2|今っ、ひ、人がっ、消えたよな!?
+SG/MAY_0049.wav|0|???
+SG/OKA_0144.wav|2|消えただろう!? 今、目の前で!
+SG/OKA_0145.wav|2|まゆりも見たか!? 見たよな!?
+SG/MAY_0050.wav|0|ん〜あ〜?
+SG/MAY_0051.wav|0|見〜て〜ない〜
+SG/OKA_0146.wav|2|見て、ない⋯⋯?
+SG/OKA_0147.wav|2|見ていない? 見ていないのか? だってついさっきまで、ここにはたくさんの人たちが歩いていたんだぞ!?
+SG/MAY_0052.wav|0|⋯⋯歩いてたかなあ?
+SG/OKA_0148.wav|2|それに店員まで消えている! こんなことはいくらなんでも有り得ない!
+SG/MAY_0053.wav|0|んー。それは仕方ないと思うよー
+SG/MAY_0054.wav|0|とにかくね、最初からこの辺には、誰もいなかったよー。あ、そっかー、オカリンは幻を見てたんだね
+SG/MAY_0055.wav|0|きっと、この暑さのせいだよー♪ トゥットゥルー♪
+SG/OKA_0150.wav|2|⋯⋯そもそも、あの人工衛星は、いったいなんだ?
+SG/OKA_0151.wav|2|まゆり、あの人工衛星だが⋯⋯
+SG/MAY_0056.wav|0|うん、びっくりしたね〜
+SG/OKA_0152.wav|2|びっくりした、だと? なにがびっくりしたのだ?
+SG/MAY_0057.wav|0|ドカーンってすごい音がしたもん
+SG/OKA_0153.wav|2|あの人工衛星は、墜ちてきたのか?
+SG/MAY_0058.wav|0|きたのかなー? 宇宙人さん乗ってるのかなー?
+SG/OKA_0154.wav|2|⋯⋯⋯⋯
+SG/MAY_0059.wav|0|ごめんなさい〜
+SG/OKA_0155.wav|2|それより制服警官さん、仮に貴方を警官Aと名付けるが、1つ聞きたいことがあるんです⋯⋯!
+SG/OKA_0156.wav|2|今ここで数千人の通行人が一瞬で消えたんです! 貴方も見ましたよね!?
+SG/OKA_0158.wav|2|そうか⋯⋯そういうことか⋯⋯!
+SG/OKA_0159.wav|2|これもすべて、機関の隠蔽工作ということだな! 警察にすら圧力をかけられるということは、この国の中枢ももはやヤツらの手の内にあるということ⋯⋯くっ、なんということだ!
+SG/OKA_0160.wav|2|だが俺の目はごまかせんぞ。いつか必ずヤツらの所業を暴き、その支配構造に終止符を打ってやる⋯⋯!
+SG/OKA_0161.wav|2|頭脳労働の後のドクターペッパーは相変わらず最高にうまいな!
+SG/MAY_0060.wav|0|オカリンは本当にドクターペッパーさんが大好きだよねー
+SG/OKA_0162.wav|2|この知的飲料の良さが分からないヤツは、人生の5分の1を損しているぞ! フゥーハハハ!
+SG/OKA_0163.wav|2|ダル。計画は順調に推移しているか
+SG/OKA_0164.wav|2|計画は計画だ。8号機の調整以外になにがあると言うのか
+SG/OKA_0165.wav|2|そろそろお前との付き合いも3年半ほどになる
+SG/OKA_0166.wav|2|細かいことはどうでもいい。それほどの付き合いの長さなのだから、いい加減俺の会話についてこられるようになってくれ
+SG/OKA_0167.wav|2|⋯⋯⋯⋯
+SG/OKA_0168.wav|2|それで、8号機の不調の原因究明は進んだか?
+SG/OKA_0169.wav|2|まゆり! まゆり! ここにバナナを持て!
+SG/MAY_0061.wav|0|⋯⋯またゲルバナ作るのー?
+SG/MAY_0062.wav|0|だってね、ゲルバナはゲルバナだもん
+SG/MAY_0063.wav|0|なんでいつもいつも、一房丸ごと入れるのー? もったいないよー
+SG/OKA_0170.wav|2|ケチケチしていては機関との戦いに勝利することなどできんぞ
+SG/MAY_0064.wav|0|勝たなくてもいいよ。あのね、バナナはまゆしぃが買ってきてるんだからねー? おかげでまゆしぃはちっともバナナが食べられません
+SG/OKA_0171.wav|2|次からは1本ずつ使うことも検討しておこう
+SG/MAY_0065.wav|0|R・E・N・G。こちらは、電話レンジ(仮)です
+SG/MAY_0066.wav|0|まゆしぃの声、聞こえてきたー?
+SG/OKA_0172.wav|2|少し黙れ。まゆしぃガイダンスが聞こえなくなる
+SG/MAY_0067.wav|0|こちらから、タイマー操作ができます
+SG/MAY_0068.wav|0|#ボタンを押した後、温めたい秒数をプッシュしてください
+SG/MAY_0069.wav|0|例えば、1分なら#60
+SG/MAY_0070.wav|0|2分なら#120⋯⋯です
+SG/OKA_0173.wav|2|なに、逆回転!?
+SG/OKA_0174.wav|2|そこに重大な意味があるかもしれない! 量子の振る舞いにも影響してくる問題であり、フントの規則を導入して––
+SG/OKA_0175.wav|2|⋯⋯ないか
+SG/OKA_0176.wav|2|⋯⋯そうか
+SG/MAY_0071.wav|0|ゲルバナのできあがり〜
+SG/OKA_0177.wav|2|ダルよ。このバナナ⋯⋯食べてみようとは思わないか? 思うはずだ。我らの理念達成の犠牲となり散ったダルに、敬礼⋯⋯!
+SG/OKA_0178.wav|2|味は関係ない。食べることに意味があるのだっ! さあダルよ、遠慮することはない。骨は拾ってやるから思い切ってずずいっと行くがいい!
+SG/OKA_0179.wav|2|⋯⋯ではまゆり。お前にその名誉を譲ろう
+SG/MAY_0072.wav|0|なんかね、ゲルバナは、中身がデロデロでぶにゅぶにゅだったよ
+SG/MAY_0073.wav|0|味もしないし、全然おいしくなかったー
+SG/OKA_0180.wav|2|デロデロでぶにゅぶにゅか⋯⋯。ダルよ、どう思う
+SG/MAY_0074.wav|0|ダルくんダルくん、鼻血出てるー
+SG/MAY_0075.wav|0|あなたのバナナ、ぶにゅぶ––
+SG/OKA_0181.wav|2|言わせるな低脳がっ!
+SG/OKA_0182.wav|2|ゲル状になったということは半固形。すなわち分子同士の結びつきが弱くなっている可能性がある
+SG/OKA_0183.wav|2|そうか、分かったぞ!
+SG/OKA_0184.wav|2|俺たちは冷凍機能だと思い込んでいたが、実は違ったのだよ!
+SG/OKA_0185.wav|2|さあお前たち、な、なんだってー!と叫ぶがいい! ここは叫ぶところだ!
+SG/MAY_0076.wav|0|冷凍の逆なら、解凍じゃないのかなー?
+SG/OKA_0186.wav|2|実に愚鈍な意見だな、まゆり! それでは普通の電子レンジと同じではないか!
+SG/MAY_0077.wav|0|じゃあ、どういうことー?
+SG/OKA_0187.wav|2|⋯⋯⋯⋯
+SG/OKA_0188.wav|2|ダルはラジ館に見物に行かないのか?
+SG/OKA_0189.wav|2|電話レンジ(仮)の件だが、俺は答えを導き出したかもしれん
+SG/OKA_0190.wav|2|そんなことは今はどうでもいい
+SG/OKA_0191.wav|2|なにを言うか。俺はいつも、この世の森羅万象すら超越したあらゆる可能性について思考を巡らせているのだ。トンデモとか言うな
+SG/OKA_0192.wav|2|ダルよ、電話レンジ(仮)は運命石の扉シュタインズゲートを開く鍵だという気がするのだが、どう思う?
+SG/OKA_0193.wav|2|あ⋯⋯!?
+SG/OKA_0194.wav|2|あ⋯⋯あ⋯⋯!
+SG/OKA_0195.wav|2|き、さま⋯⋯
+SG/OKA_0196.wav|2|貴様は、死んだはずだ! なぜ、ここに⋯⋯!?
+SG/OKA_0197.wav|2|しかも––
+SG/OKA_0198.wav|2|無傷⋯⋯!
+SG/OKA_0199.wav|2|無事だったのか? ケガは平気なのか? いや、そんなはずはない、牧瀬紅莉栖は何者かに刺されて血まみれで––
+SG/OKA_0200.wav|2|またその話、とは、どういう意味だ?
+SG/OKA_0201.wav|2|メール? 俺が?
+SG/OKA_0202.wav|2|なにをバカな! 牧瀬紅莉栖が殺されているのを見たのは、ほんの3時間前だぞ!
+SG/OKA_0203.wav|2|ネットで妙な考察サイトでも見たのか、ダル。お前がトンデモ理論を言い出すとは珍しい
+SG/OKA_0204.wav|2|これは⋯⋯3時間前にダルに送ったメールだ
+SG/OKA_0205.wav|2|ある⋯⋯。実体が、ある。やはり幽霊だというのは考えすぎか⋯⋯
+SG/OKA_0206.wav|2|⋯⋯俺は真実を知りたいだけだ
+SG/OKA_0207.wav|2|俺は確かに見たのだ!
+SG/OKA_0208.wav|2|この低脳めが! そうではない!
+SG/OKA_0210.wav|2|⋯⋯中止!?
+SG/OKA_0211.wav|2|わ、我が名は鳳凰院凶真だ
+SG/OKA_0212.wav|2|行く、とはどういう意味だ?
+SG/OKA_0213.wav|2|講義をする方だったのか⋯⋯
+SG/OKA_0214.wav|2|ほう、タイムマシンか⋯⋯
+SG/OKA_0215.wav|2|異議あり!
+SG/OKA_0216.wav|2|タイムマシンが作れないと決めつけるのは早計だ
+SG/OKA_0217.wav|2|⋯⋯⋯⋯
+SG/OKA_0218.wav|2|では12番目の理論が発見されたとしたらどうかな?
+SG/OKA_0219.wav|2|⋯⋯っ
+SG/OKA_0220.wav|2|空間に開いた、抜け道のようなもの⋯⋯だろう?
+SG/OKA_0221.wav|2|タイムパラドックス⋯⋯質量保存の法則?
+SG/OKA_0222.wav|2|⋯⋯⋯⋯
+SG/MAY_0078.wav|0|あれれ? オカリンだー。トゥットゥルー♪
+SG/OKA_0223.wav|2|ルカ子よ、お前、俺が与えた刀はどうした
+SG/OKA_0224.wav|2|そうだ。あれはお前の力を制御するために買ってやったのだぞ
+SG/MAY_0079.wav|0|あー、アキバの武器屋本舗で買ったやつでしょー? 980円だっけ––
+SG/OKA_0225.wav|2|まゆり! それ以上言うとヤツらに消されるぞ! この件については口を出すな!
+SG/MAY_0080.wav|0|え、消されちゃうのー? オカリン、心配してくれてありがとうー。でもでも、ヤツらって誰かな?
+SG/OKA_0226.wav|2|それでルカ子よ。妖刀・五月雨はちゃんと使っているのか
+SG/OKA_0227.wav|2|あれを持ち清心斬魔せいしんざんま流を極めさえすれば、お前は己の内にある邪悪な炎に焼かれずに済む
+SG/OKA_0228.wav|2|俺は岡部ではない
+SG/MAY_0081.wav|0|オカリンだよー
+SG/OKA_0229.wav|2|分かればいいのだ。では合い言葉を
+SG/OKA_0230.wav|2|違う! コンガリゥではなく、コングルゥだ!
+SG/MAY_0082.wav|0|美しい師弟関係だねー。えっへへー。まゆしぃは腐女子じゃないけど、ちょっとドキドキしてきちゃうよー
+SG/OKA_0231.wav|2|それで、なぜまゆりはここにいる?
+SG/MAY_0083.wav|0|るかくんに会いに来たんだよー
+SG/MAY_0084.wav|0|来月のコミマで、雷ネットのキラリちゃんコスをしてほしいってずっと頼んでるのに、ちっともOKしてくれないんだよー
+SG/MAY_0085.wav|0|でもでもー、るかくんは絶対似合うと思うんだー
+SG/MAY_0086.wav|0|こんなかわいい子が女の子のはずがないって大人気になるよ? ね? しようよー、コスプレデビュー
+SG/OKA_0232.wav|2|そんな下らないことは後でやるんだな
+SG/MAY_0087.wav|0|えー? まゆしぃにとっては大事なことだもん
+SG/OKA_0233.wav|2|俺にとっては下らないことなのだ!
+SG/OKA_0234.wav|2|それよりルカ子よ、俺がこの神社を訪ねたのは他でもない。お祓いを頼みたいのだが、やってもらえないだろうか
+SG/OKA_0235.wav|2|いや、そこまで大げさにしなくていい。気休めでいいのだ
+SG/OKA_0236.wav|2|というわけで、例のアレを持ってこい
+SG/OKA_0237.wav|2|違う! お祓いに妖刀は必要ないだろう! お祓いと言ったらアレに決まっている!
+SG/OKA_0238.wav|2|正式な名称は分からんが、棒に白い紙がフサフサと付いていて、神主がワサワサと振るやつだ!
+SG/MAY_0088.wav|0|あはは、今のオカリンの説明、ゆとりっぽいねー♪
+SG/MAY_0089.wav|0|んじゃ、まゆしぃはこれからバイトだから、もう行くねー
+SG/OKA_0239.wav|2|そうか。頑張ってこい。バイトが終わったら直接帰るのか?
+SG/MAY_0090.wav|0|うん

filelists/SG_val.txt.cleaned

@@ -0,0 +1,300 @@
+SG/MAY_0000.wav|0|neenee. nanibutsubutsu iclterunoo?
+SG/MAY_0001.wav|0|okariN? neecltebaa.
+SG/MAY_0002.wav|0|darekato deNwachuu?
+SG/OKA_0027.wav|2|...... iya, kochirano hanashida. moNdainai, koreyori kaijooni seNnyuu suru.
+SG/OKA_0028.wav|2|aa, dokUtaachuubachi nakabachiwa nukegakeo shIta. taclpurito sono kaNgaeni tsuite kIkasete morau tsumorisa.
+SG/OKA_0029.wav|2|...... nani!? kikaNga ugokidashIte irudato!?
+SG/OKA_0030.wav|2|sooka, sorega uNmeesekino tobirashUtaiNzugeetono seNtakUka. erupusaikoNguruu.
+SG/MAY_0003.wav|0|saclki, keetaide dareto hanashItetanoo?
+SG/OKA_0031.wav|2|kIkuna. sorega mayurino tamedemo aru.
+SG/MAY_0004.wav|0|soo naN daa. okariN, arigatoo.
+SG/MAY_0005.wav|0|soreyori okariNokariN.
+SG/OKA_0032.wav|2|mayuriyo, itsumo iclte irudaroo. oreno kotoo okariNto yobunato.
+SG/MAY_0006.wav|0|ee? demo mukashIkara soo yoNdetayo?
+SG/OKA_0033.wav|2|sorewa mukashino hanashida. imano orewa hooooiN hooooiN kyoo makyooma. sekaijuuno himitsUsoshIkikara nerawareru, kyookino macldo saieNtisUtoda. fuuu hahaha!
+SG/MAY_0007.wav|0|daclte, muzukashikUte oboerarenaishi.
+SG/MAY_0008.wav|0|soreni, okabe okabe riNtaroo riNtarooto ichimojimo acltenaiyoo? okashiine, ecle hee.
+SG/MAY_0009.wav|0|dene, okariN. eclto, oshietehoshii Ndakedo.
+SG/MAY_0010.wav|0|korekara kokode, naniga hajimarunoo?
+SG/OKA_0034.wav|2|omaewa, soremo shirazu kokomade oreni tsuite kitato iu noka.
+SG/MAY_0011.wav|0|uN.
+SG/OKA_0035.wav|2|korekara kokode hajimaru nowa, dokUtaachuu hachino kIshakaikeNda.
+SG/MAY_0012.wav|0|kIshakaikeN? demoo, kishasaNnaNte miataranai kiga suruyo?
+SG/OKA_0036.wav|2|aruiwa, kikaNni yoru naNrakano boogaio uketa nokamo shirenaina.
+SG/OKA_0037.wav|2|makIkomareru nowa, kaNbeNdagana.
+SG/MAY_0013.wav|0|makimakI tokage? a, soreo iunara eri makI tokagedanee. ecle hee.
+SG/OKA_0038.wav|2|mayuri, kio tsUkero. osorakU kono kIshakaikeN, naNrakano jikeNga oki--
+SG/MAY_0014.wav|0|jishiNkanaa? shiNdoni? magunichuudoni? shiNdoto magunichuudoclte doo chigau Ndaclkee......
+SG/OKA_0039.wav|2|bakUhatsu...... dato!?
+SG/OKA_0040.wav|2|naNda...... korewa?
+SG/OKA_0041.wav|2|korewa niou. iNboono nioida. nanio kakUshitai Nda? saclkino bakUhatsuwa naNda?
+SG/OKA_0042.wav|2|oreda. doomo iyana yokaNga suru. oretachiga shiranai tokorode nanikaga okoclte iru yooda.
+SG/OKA_0043.wav|2|...... aa, wakaclteru. muchawa shinaisa. oremo inochiwa oshiikarana. erupusaikoNguruu.
+SG/OKA_0044.wav|2|mayuri, nanio shIte iru?
+SG/MAY_0015.wav|0|NN?
+SG/MAY_0016.wav|0|anone, uu paga hoshiinaaclte.
+SG/OKA_0045.wav|2|yareba ii. uu paga ataruka dookawa hoshoo dekinaigana.
+SG/MAY_0017.wav|0|demone, mayu shiiwa ima, hyakueNdamao kirashIchaclte iru nodesU.
+SG/MAY_0018.wav|0|dakara, okariNokariN, hyakueN kashitee?
+SG/OKA_0046.wav|2|amacltareruna mayuri. kiNwa kasaN. orega omaeni jiNseeno kibishisao oshiete yaru.
+SG/MAY_0019.wav|0|a, aaa......
+SG/MAY_0020.wav|0|acl, uu padayo. shIkamo metaru. metaruuu p a.
+SG/OKA_0047.wav|2|sorewa reana noka?
+SG/MAY_0021.wav|0|sugoku!
+SG/OKA_0048.wav|2|fuN, dewa mayurini kurete yaroo.
+SG/MAY_0022.wav|0|hoN too? iino? okariN.
+SG/OKA_0049.wav|2|hooooiN kyoosanada.
+SG/MAY_0023.wav|0|ecle hee, arigatoookariN.
+SG/OKA_0050.wav|2|............
+SG/OKA_0051.wav|2|dooyara hajimaru yoodana.
+SG/OKA_0052.wav|2|ikuzo mayuri.
+SG/MAY_0024.wav|0|NN, maclte maclte. namae kakanakUcha.
+SG/MAY_0025.wav|0|taimumashiN? ano hItoga tsUkucltanoo?
+SG/OKA_0053.wav|2|dooookuuuutaaaaacl!
+SG/OKA_0054.wav|2|bakani surunimohodoga aruzo!
+SG/OKA_0055.wav|2|orega darena nokawa doodemo ii! soreyori, ima anataga kataclta taimumashiNno riroNwa icltai naNda!?
+SG/OKA_0056.wav|2|joNtaitaano pakuridewanaika! anatawa soredemo hatsumeeka kacl!
+SG/OKA_0057.wav|2|dete iku nowa anatada, dokUtaa! hajio shire! koNriNzai koNriNzai, anataniwa hatsumeekao nanoru shIkakuwa naizo cl!
+SG/OKA_0058.wav|2|w a, n a, secl...... NN?
+SG/OKA_0059.wav|2|a......
+SG/OKA_0060.wav|2|k i, kIsama, kikaNno niNgeNka!?
+SG/OKA_0061.wav|2|kucl, masaka kokomade tega mawaclte irutowa....... oreto shIta kotoga.
+SG/OKA_0062.wav|2|............
+SG/OKA_0063.wav|2|kokode oreni nanika sureba hItomeni tsukuzo. soo nareba kIsamamo iroiro mazuidaroo.
+SG/OKA_0064.wav|2|soreni kotaeru giriwa nai. kikaNno yarikatawa wakaclte iru.
+SG/OKA_0065.wav|2|oreda. kikaNno eejeNtoni tsUkamaclta....... aa, makIse kurenairiisumikada, ano oNnaniwa kio tsUkero...... iya moNdainai, kokowa naNtoka kirinuke--
+SG/OKA_0066.wav|2|kucl, nanio suru!
+SG/OKA_0067.wav|2|............
+SG/OKA_0068.wav|2|k i, kIsamani kotaeru giriwa naiga ichioo oshiete yaroo. sorewa oreigaiga sawaruto jidootekini deNgeNga ofuni naru, tokubetsUseeno keetaina nodacl. fuuu hahaha!
+SG/OKA_0069.wav|2|...... cl.
+SG/OKA_0070.wav|2|saclkItowa itsuno kotoda?
+SG/OKA_0071.wav|2|orewa subete omitooshina noda.
+SG/OKA_0072.wav|2|teNsaishoojoyo, tsugi au tokiwa katakidooshidana!
+SG/OKA_0073.wav|2|sarabada, fuuu hahaha!
+SG/OKA_0074.wav|2|k i, kIkaNme, arehodono eejeNtoo okurikoNde kurutowa, tsuini hoNkini naclta yoodana......!
+SG/OKA_0075.wav|2|d a, daga, orewa mada yatsurani tsUkamaru wakeniwa ikaN noda......
+SG/OKA_0076.wav|2|chiicl! mayurio oite kita......!
+SG/OKA_0077.wav|2|N......? meeruka?
+SG/OKA_0078.wav|2|......?
+SG/OKA_0079.wav|2|kucl, mayuri. naze denai.
+SG/OKA_0080.wav|2|hacl, masaka makIse kurenairii sume, mayurio saracltana......!
+SG/OKA_0081.wav|2|onoreee. sorega kikaNno yarikataka......!
+SG/OKA_0082.wav|2|modorushIkanaika......
+SG/OKA_0083.wav|2|kUkuku, ano oNna, oreni ojikizuitaka.
+SG/OKA_0084.wav|2|yokarou. ima iwakuno tokorowa minogashIte yaruto shiyoo.
+SG/OKA_0085.wav|2|mayuri, naze deNwani denai. sorosoro kaeruzo.
+SG/MAY_0026.wav|0|a, okariN. metaruuu paga inaku naclchaclta.
+SG/OKA_0086.wav|2|inaku naclta? kaclteni ugokidashIta noka. sorewa jitsuni faNtajiidana.
+SG/MAY_0027.wav|0|otoshita mitai......
+SG/OKA_0087.wav|2|mitsUkaranainara akiramero. mata atereba ii.
+SG/MAY_0028.wav|0|ataricl konaiyo. dacltene, metaruuu pawa necltoookUshoNde, ichimaNeN chIkai puremiaga tsuiteru Ndayo?
+SG/OKA_0088.wav|2|n a...... n i......?
+SG/OKA_0089.wav|2|mayuriyo, icltai dokode otoshita noda!?
+SG/MAY_0029.wav|0|wakaNnaikara sagashIteru Nda you....... ato, mitsUketemo, uranaikaranee?
+SG/OKA_0090.wav|2|fuhaha, sono ichimaNeN, kono oreno keNkyuushikiNni shIteyaru.
+SG/MAY_0030.wav|0|dakara, uranaicltebaa. mayu shiino namae kaichacltashi.
+SG/MAY_0031.wav|0|tuclturuu uu p a saNuu pasaN, dete oi dee.
+SG/OKA_0091.wav|2|onore, kiNnishIka kyoomino nai geretsuna yatsume, hajio shire......!
+SG/MAY_0032.wav|0|okariNmonee.
+SG/OKA_0093.wav|2|!?
+SG/MAY_0033.wav|0|himee...... kana?
+SG/OKA_0094.wav|2|mayuri, kokoni iro.
+SG/OKA_0095.wav|2|hicl......
+SG/OKA_0096.wav|2|e, n a, naNde......?
+SG/MAY_0034.wav|0|okariN, doo shItano......?
+SG/OKA_0097.wav|2|d e, deruzo cl.
+SG/OKA_0098.wav|2|haa, haa......
+SG/MAY_0035.wav|0|neenee, naniga acltanoo? kaoiro, sugokuwaruikedo......
+SG/OKA_0099.wav|2|hItoga...... shiNdeta.
+SG/MAY_0036.wav|0|e......
+SG/OKA_0100.wav|2|-- cl.
+SG/OKA_0101.wav|2|kieta......
+SG/OKA_0102.wav|2|oi, sokono kIsama. oretachiga miete iruka?
+SG/OKA_0103.wav|2|...... naze nanimo kotaenai. kIsamani kiite iru Ndazo? monitano soclchigawani iru, kIsamanida.
+SG/OKA_0104.wav|2|fuN. manukemeNo shioclte. tsumaraN yatsuda.
+SG/OKA_0105.wav|2|kIsamakaradato, oretachiwa terebino monitano nakani iru yooni mierudaroona. kUkuku, daga sorewa ookina machigaida.
+SG/OKA_0106.wav|2|monitano nakani iru nowa kIsamana nodayo. kIsamaga geNjitsudato omoclte iru sono sekaiwa, jitsuwa subete kyokoo. mochiroN kIsamajishiNmona.
+SG/OKA_0107.wav|2|shiNno geNjitsu, sorewa kochiragawani aru.
+SG/OKA_0108.wav|2|jibuNga nanio shItekI sarete iru nokasura wakaclte inaika. murimo nai.
+SG/OKA_0109.wav|2|maa ii. kIsamaniwa wakariyasUku, oretachino kotoni tsuite setsumee shIte yaroodewanaika.
+SG/OKA_0110.wav|2|mazu, oretachiga iru nowa dokokato iu kotoda. kokowa tookyoo, akIhabarani aru, miraigajecltokeNkyuujoda. oretachiwa fudaNraboto yoNde iru. sekaino shIhaikoozooo tsUkurikaeruto iu waga yaboono kyoteNdana.
+SG/MAY_0037.wav|0|soo naN daa. warui koto shIcha damena Ndayo, okariN.
+SG/OKA_0111.wav|2|mayuriwa sUkoshi damaclte iro.
+SG/OKA_0112.wav|2|ekIkara detara chuuoodoorio susumi, suehirochooekino koosateNo kuramaebashidoorie sasetsu. tsugino shiNgoono iclpo temaeno rojio hairuto, daihiyama biruto iu furukUsai zaclkyobiruga aru. sono nikaini waga rabowa kyoo kamaete iru.
+SG/OKA_0113.wav|2|mejirushiwa, biru iclkaini aru burauNkaNkoobooto iu maniaclkuna tenaNtodana.
+SG/OKA_0114.wav|2|imadoki, kyuushIkino burauNkaNterebidakeo atsUkaclte iru shoclpuda. ikani deNkigaidearu akIhabarato iedomo juyooga arutowa omoenai, sabireta misena noda.
+SG/OKA_0115.wav|2|daga burauNkaNkoobooteNchoodearu teNnoojiteNnoojito iu oclsaNwa, kono biruno oonaademo aru.
+SG/OKA_0116.wav|2|yueni, imaya kyuupiclchide toshIkaihatsuga susumi, chikamo kootoo shIteru kono akIhabaradearooto, doorakumarudashino miseo kamaete irareruto iu wakeda.
+SG/OKA_0117.wav|2|saiwainimo ano teNchoowa, hItoo miru mewa aru yoodena. kono oreno karisumaseeo minuki, biru nikaio marumaru waNfuroa, tadadoozeNde kashIte kureta noda. fuuu hahaha!
+SG/OKA_0118.wav|2|daga miraigajecltokeNkyuujowa shiNkokuna jiNzaifusokuno tame, yuushuuna keNkyuuiNo zuiji boshuuchuuda. imano tokoro, shozokUkeNkyuuiNwa--
+SG/MAY_0038.wav|0|okariNokariN, sokowa rabomeNclte iwana kyaa. shozokUkeNkyuuiNjanakUte.
+SG/OKA_0119.wav|2|...... rabomeN, sunawachi raboratoriimeNbaawa, oreo fUkumete saNniNdearu.
+SG/OKA_0120.wav|2|rabomeNnaNbaa zerozero ichi, rabosoosetsUshani shIte kyookino macldo saieNtisUtodearu kono ore, hooooiN kyooshiN.
+SG/MAY_0039.wav|0|okariNclte yobikatano hooga kawaiinonii.
+SG/OKA_0121.wav|2|soshIte kosUpurega shumino kooiclteN, rabomeNnaNbaa zerozero nii, shiina mayuri.
+SG/MAY_0040.wav|0|tuclturuu mayu shiideesU. kiru NjanakUte tsUkuru noga shumidayo.
+SG/OKA_0122.wav|2|saigoni suupaahakaa, rabomeNnaNbaa zerozero saN, hashidawa shida itari itaruda.
+SG/OKA_0123.wav|2|soNna wareware saNniNde koosee sareru miraigajecltokeNkyuujono katsudoonaiyoowa, sonomono zubari hatsumeedearu.
+SG/OKA_0124.wav|2|shoosaiwa waga rabono hoomupeejio mite kure.
+SG/OKA_0125.wav|2|mochiroN yamino shIhaikeNryokUto tatakau tameno miraigajecltoga saiyuuseNjikoodaga, sono keNkyuukara hasee suru fUkusaNbutsutekina hatsumeemo ooi. t o iuka imano tokorowa soclchibakarida.
+SG/OKA_0126.wav|2|sudeni warewarewa yacltsuno miraigajecltoo kaNsee saseta. daga korewa mada joshoodeshIka nai. miraigajecltono aideawa, oreno nakani hyaku hachimadearu noda.
+SG/OKA_0127.wav|2|hItono boNnoono suuto ie, kono@ chaNnerunoome.
+SG/OKA_0128.wav|2|soreto, orega hanashIte iru nodakara kUchidashio surunato iclte irudaroo.
+SG/OKA_0129.wav|2|hItorigotodewa nai. mite wakaranai noka. orewa ima, monitano mukooni iru koitsuni hanashIkakete iru noda.
+SG/MAY_0041.wav|0|a, ima sono hIto, niyariclte waracltaa.
+SG/OKA_0130.wav|2|onorekisama, nanio waraclte iru noka! monitano nakano soNzaino kUseni!
+SG/MAY_0042.wav|0|tsuujinai Njanaikanaa?
+SG/OKA_0131.wav|2|oretachini hanashIkakerarete iru kotonisura, kizuite inairashiina. jikakuga naito iu nowa, jitsuni fUkoona kotoda.
+SG/MAY_0043.wav|0|sono hItoni shIte mitara, mayu shiitachiga geemumitaini mieteru nokanaa?
+SG/MAY_0044.wav|0|Nja, darukuNno daisUkina nijigeNno oNnanokotachimo soonanoo?
+SG/OKA_0132.wav|2|daruno yomeno hanashiwa doodemo ii.
+SG/OKA_0133.wav|2|naina.
+SG/OKA_0134.wav|2|yueni, sono yoona giroNwa fumoo. sekaino shIhaikoozooo uchIkudakU hoohooni tsuite kaNgaeru hooga, yohodo yuuigida.
+SG/OKA_0135.wav|2|damare suupaahakaa. orewa chuunibyoodewa nai.
+SG/OKA_0136.wav|2|hooooiN...... kyooshiNdacl!
+SG/OKA_0137.wav|2|yareyare. daruno, hItotono komyunikeeshoNno torenasawa, doredake tacltemo naoranaina.
+SG/MAY_0045.wav|0|au....... hari, yubini sasaclta......
+SG/MAY_0046.wav|0|mayu shiiwa okariNno hItojichidakara, kokoni iyooto omoimaasU.
+SG/OKA_0138.wav|2|iya, zeNzeN.
+SG/OKA_0139.wav|2|kono buaisoo arupakamega.
+SG/OKA_0140.wav|2|N?
+SG/MAY_0047.wav|0|kicltone, arupakasaNga, okoclchaclta Ndayoo.
+SG/OKA_0141.wav|2|kucl. atode shuuri dekiruka kIkini ikanakereba.
+SG/OKA_0142.wav|2|kieta......
+SG/MAY_0048.wav|0|dooka shItaa?
+SG/OKA_0143.wav|2|imacl, h i, hItogacl, kietayona!?
+SG/MAY_0049.wav|0|???
+SG/OKA_0144.wav|2|kietadaroo!? ima, menomaede!
+SG/OKA_0145.wav|2|mayurimo mitaka!? mitayona!?
+SG/MAY_0050.wav|0|N~ a~?
+SG/MAY_0051.wav|0|m i~ t e~ nai~
+SG/OKA_0146.wav|2|mite, nai......?
+SG/OKA_0147.wav|2|mite inai? mite inai noka? daclte tsui saclkimade, kokoniwa takUsaNno hItotachiga aruite ita Ndazo!?
+SG/MAY_0052.wav|0|...... aruitetakanaa?
+SG/OKA_0148.wav|2|soreni teNiNmade kiete iru! koNna kotowa ikura naNdemo arienai!
+SG/MAY_0053.wav|0|NN. sorewa shIkatanaito omouyoo.
+SG/MAY_0054.wav|0|tonikakune, saishokara kono atariniwa, daremo inakacltayoo. a, soclkaa, okariNwa maboroshio mite taNdane.
+SG/MAY_0055.wav|0|kiclto, kono atsusano seidayoo tuclturuu.
+SG/OKA_0150.wav|2|...... somosomo, ano jiNkooeeseewa, icltai naNda?
+SG/OKA_0151.wav|2|mayuri, ano jiNkooeeseedaga......
+SG/MAY_0056.wav|0|uN, biclkuri shItane~
+SG/OKA_0152.wav|2|biclkuri shIta, dato? naniga biclkuri shIta noda?
+SG/MAY_0057.wav|0|dokaaNclte sugoi otoga shItamoN.
+SG/OKA_0153.wav|2|ano jiNkooeeseewa, ochite kita noka?
+SG/MAY_0058.wav|0|kita nokanaa? uchuujiNsaN noclteru nokanaa?
+SG/OKA_0154.wav|2|............
+SG/MAY_0059.wav|0|gomeNnasai~
+SG/OKA_0155.wav|2|soreyori seefUkukeekaNsaN, karini anatao keekaN ei t o nazukeruga, hItotsU kikItai kotoga aru NdesU......!
+SG/OKA_0156.wav|2|ima kokode suuseNniNno tsuukooniNga iclshuNde kieta NdesU! anatamo mimashItayone!?
+SG/OKA_0158.wav|2|sooka...... sooyuu kotoka......!
+SG/OKA_0159.wav|2|koremo subete, kikaNno iNpeekoosakUto iu kotodana! keesatsunisura atsuryokuo kakerareruto iu kotowa, kono kunino chuusuumo mohaya yatsurano tenouchini aruto iu koto...... kucl, naNto iu kotoda!
+SG/OKA_0160.wav|2|daga oreno mewa gomakaseNzo. itsUka kanarazu yatsurano shogyooo abaki, sono shIhaikoozooni shuushifuo uclte yaru......!
+SG/OKA_0161.wav|2|zunooroodoono atono dokUtaapeclpaawa aikawarazu saikooni umaina!
+SG/MAY_0060.wav|0|okariNwa hoNtooni dokUtaapeclpaasaNga daisUkidayonee.
+SG/OKA_0162.wav|2|kono chItekiiNryoono yosaga wakaranai yatsuwa, jiNseeno gofuNno ichio soN shIte iruzo! fuuu hahaha!
+SG/OKA_0163.wav|2|daru. keekakuwa juNchooni suii shIte iruka.
+SG/OKA_0164.wav|2|keekakuwa keekakuda. hachigookino chooseeigaini naniga aruto iu noka.
+SG/OKA_0165.wav|2|sorosoro omaetono tsUkiaimo saNneN haNhodoni naru.
+SG/OKA_0166.wav|2|komakai kotowa doodemo ii. sorehodono tsUkiaino nagasana nodakara, iikageN oreno kaiwani tsuite korareru yooni naclte kure.
+SG/OKA_0167.wav|2|............
+SG/OKA_0168.wav|2|sorede, hachigookino fUchoono geNiNkyuumeewa susuNdaka?
+SG/OKA_0169.wav|2|mayuri! mayuri! kokoni bananao mote!
+SG/MAY_0061.wav|0|...... mata gerubana tsUkurunoo?
+SG/MAY_0062.wav|0|dacltene, gerubanawa gerubanadamoN.
+SG/MAY_0063.wav|0|naNde itsumo itsumo, ichiboo marugoto irerunoo? mocltainaiyoo.
+SG/OKA_0170.wav|2|kechIkechi shIte itewa kikaNtono tatakaini shoori suru kotonado dekiNzo.
+SG/MAY_0064.wav|0|katanakUtemo iiyo. anone, bananawa mayu shiiga kaclte kIteru Ndakaranee? okagede mayu shiiwa chicltomo bananaga taberaremaseN.
+SG/OKA_0171.wav|2|tsugikarawa iclpoNzutsU tsukau kotomo keNtoo shIte okoo.
+SG/MAY_0065.wav|0|aaruiienujii. kochirawa, deNwareNji( kari) desU.
+SG/MAY_0066.wav|0|mayu shiino koe, kIkoete kitaa?
+SG/OKA_0172.wav|2|sUkoshi damare. mayu shiigaidaNsuga kIkoenaku naru.
+SG/MAY_0067.wav|0|kochirakara, taimaasoosaga dekimasU.
+SG/MAY_0068.wav|0|# botaNo oshIta nochi, atatametai byoosuuo puclshu shIte kudasai.
+SG/MAY_0069.wav|0|tatoeba, iclpuNnara# rokujuu.
+SG/MAY_0070.wav|0|nifuNnara# hyaku nijuu...... desU.
+SG/OKA_0173.wav|2|nani, gyakUkaiteN!?
+SG/OKA_0174.wav|2|sokoni juudaina imiga arukamo shirenai! ryooshino furumainimo eekyoo shIte kuru moNdaideari, fuNtono kisokuo doonyuu shIte--
+SG/OKA_0175.wav|2|...... naika.
+SG/OKA_0176.wav|2|...... sooka.
+SG/MAY_0071.wav|0|gerubanano dekiagari~
+SG/OKA_0177.wav|2|daruyo. kono banana...... tabete miyootowa omowanaika? omou hazuda. warerano rineNtaclseeno giseeto nari chiclta daruni, keeree......!
+SG/OKA_0178.wav|2|ajiwa kaNkee nai. taberu kotoni imiga aru nodacl! saadaruyo, eNryo suru kotowa nai. honewa hiroclte yarukara omoikiclte zuzu iclto ikuga ii!
+SG/OKA_0179.wav|2|...... dewa mayuri. omaeni sono meeyoo yuzuroo.
+SG/MAY_0072.wav|0|naNkane, gerubanawa, nakamiga derodero debuni yubunyudacltayo.
+SG/MAY_0073.wav|0|ajimo shinaishi, zeNzeN oishikunakacltaa.
+SG/OKA_0180.wav|2|derodero debuni yubunyuka....... daruyo, doo omou.
+SG/MAY_0074.wav|0|darukuN darukuN, hanaji deteruu.
+SG/MAY_0075.wav|0|anatano banana, bunyubu--
+SG/OKA_0181.wav|2|iwaseruna teenoogacl!
+SG/OKA_0182.wav|2|gerujooni nacltato iu kotowa haNkokee. sunawachi buNshidooshino musubitsUkiga yowakunaclte iru kanooseega aru.
+SG/OKA_0183.wav|2|sooka, wakacltazo!
+SG/OKA_0184.wav|2|oretachiwa reetookinoodato omoikoNde itaga, jitsuwa chigaclta nodayo!
+SG/OKA_0185.wav|2|saaomaetachi, n a, naNdacltee! t o sakebuga ii! kokowa sakebu tokoroda!
+SG/MAY_0076.wav|0|reetoono gyakunara, kaitoojanai nokanaa?
+SG/OKA_0186.wav|2|jitsuni gudoNna ikeNdana, mayuri! soredewa fUtsuuno deNshireNjito onajidewanaika!
+SG/MAY_0077.wav|0|jaa, dooyuu kotoo?
+SG/OKA_0187.wav|2|............
+SG/OKA_0188.wav|2|daruwa rajikaNni keNbutsuni ikanai noka?
+SG/OKA_0189.wav|2|deNwareNji( kari) n o keNdaga, orewa kotaeo michibikidashItakamo shireN.
+SG/OKA_0190.wav|2|soNna kotowa imawa doodemo ii.
+SG/OKA_0191.wav|2|nanio iuka. orewa itsumo, kono yono shiNrabaNshoosura chooetsu shIta arayuru kanooseeni tsuite shIkooo megurasete iru noda. toNdemotoka iuna.
+SG/OKA_0192.wav|2|daruyo, deNwareNji( kari) w a uNmeesekino tobirashUtaiNzugeetoo hirakU kagidato iu kiga suru nodaga, doo omou?
+SG/OKA_0193.wav|2|a......!?
+SG/OKA_0194.wav|2|a...... a......!
+SG/OKA_0195.wav|2|k i, sama......
+SG/OKA_0196.wav|2|kIsamawa, shiNda hazuda! naze, kokoni......!?
+SG/OKA_0197.wav|2|shIkamo--
+SG/OKA_0198.wav|2|mukizu......!
+SG/OKA_0199.wav|2|bujidaclta noka? kegawa heekina noka? iya, soNna hazuwa nai, makIse kurenairiisumikawa nanimonokani sasarete chimamirede--
+SG/OKA_0200.wav|2|mata sono hanashi, towa, dooyuu imida?
+SG/OKA_0201.wav|2|meeru? orega?
+SG/OKA_0202.wav|2|nanio bakana! makIse kurenairiisumikaga korosarete iru noo mita nowa, hoNno saNjikaN maedazo!
+SG/OKA_0203.wav|2|necltode myoona koosatsusaitodemo mita noka, daru. omaega toNdemoriroNo iidasUtowa mezurashii.
+SG/OKA_0204.wav|2|korewa...... saNjikaN maeni daruni okuclta meeruda.
+SG/OKA_0205.wav|2|aru....... jicltaiga, aru. yahari yuureedato iu nowa kaNgaesugika......
+SG/OKA_0206.wav|2|...... orewa shiNjitsuo shiritaidakeda.
+SG/OKA_0207.wav|2|orewa tashIkani mita noda!
+SG/OKA_0208.wav|2|kono teenoomega! soodewa nai!
+SG/OKA_0210.wav|2|...... chuushi!?
+SG/OKA_0211.wav|2|w a, waga nawa hooooiN kyoosanada.
+SG/OKA_0212.wav|2|iku, towa dooyuu imida?
+SG/OKA_0213.wav|2|koogio suru hoodaclta noka......
+SG/OKA_0214.wav|2|hoo, taimumashiNka......
+SG/OKA_0215.wav|2|igi ari!
+SG/OKA_0216.wav|2|taimumashiNga tsUkurenaito kimetsUkeru nowa sookeeda.
+SG/OKA_0217.wav|2|............
+SG/OKA_0218.wav|2|dewa juu nibaNmeno riroNga haclkeN saretato shItara dookana?
+SG/OKA_0219.wav|2|...... cl.
+SG/OKA_0220.wav|2|kuukaNni hiraita, nukemichino yoona mono...... daroo?
+SG/OKA_0221.wav|2|taimuparadoclkUsu...... shItsuryoohozoNno hoosoku?
+SG/OKA_0222.wav|2|............
+SG/MAY_0078.wav|0|arere? okariN daa. tuclturuu.
+SG/OKA_0223.wav|2|rukakoyo, omae, orega ataeta katanawa doo shIta.
+SG/OKA_0224.wav|2|sooda. arewa omaeno chIkarao seegyo suru tameni kaclte yaclta nodazo.
+SG/MAY_0079.wav|0|aa, akibano bukiya hoNpode kaclta yatsudeshoo? kyuuhyakU hachijuueNdaclke--
+SG/OKA_0225.wav|2|mayuri! soreijoo iuto yatsurani kesareruzo! kono keNni tsuitewa kUchio dasuna!
+SG/MAY_0080.wav|0|e, kesarechaunoo? okariN, shiNpai shIte kurete arigatooo. demo demo, yatsuraclte darekana?
+SG/OKA_0226.wav|2|sorede rukakoyo. yootoosamidarewa chaNto tsUkaclte iru noka.
+SG/OKA_0227.wav|2|areo mochi seeshiNzaNma sei shiNzaNmaryuuo kimesae sureba, omaewa onoreno uchini aru jaakuna honooni yakarezuni sumu.
+SG/OKA_0228.wav|2|orewa okabedewa nai.
+SG/MAY_0081.wav|0|okariNdayoo.
+SG/OKA_0229.wav|2|wakareba ii noda. dewa aikotobao.
+SG/OKA_0230.wav|2|chigau! koNgari udewanaku, koNguruuda!
+SG/MAY_0082.wav|0|utsUkushii shIteekaNkeedanee. ecle hee. mayu shiiwa kUsa joshijanaikedo, choclto dokidoki shIte kIchauyoo.
+SG/OKA_0231.wav|2|sorede, naze mayuriwa kokoni iru?
+SG/MAY_0083.wav|0|rukakuNni aini kita Ndayoo.
+SG/MAY_0084.wav|0|raigetsuno komimade, kaminarinecltono kirari chaN kosuo shItehoshiiclte zuclto tanoNderunoni, chicltomo ookei shIte kurenai Ndayoo.
+SG/MAY_0085.wav|0|demo demoo, rukakuNwa zecltai niauto omou N daa.
+SG/MAY_0086.wav|0|koNna kawaii koga oNnanokono hazuga naiclte dainiNkini naruyo? n e? shiyooyoo, kosUpuredebyuu.
+SG/OKA_0232.wav|2|soNna kudaranai kotowa atode yaru Ndana.
+SG/MAY_0087.wav|0|ee? mayu shiini tocltewa daijina kotodamoN.
+SG/OKA_0233.wav|2|oreni tocltewa kudaranai kotona noda!
+SG/OKA_0234.wav|2|soreyori rukakoyo, orega kono jiNjao tazuneta nowa hokademo nai. oharaio tanomitai nodaga, yaclte moraenaidarooka.
+SG/OKA_0235.wav|2|iya, sokomade oogesani shinakUteii. kiyasumede ii noda.
+SG/OKA_0236.wav|2|t o iu wakede, reeno areo moclte koi.
+SG/OKA_0237.wav|2|chigau! oharaini yootoowa hItsuyoo naidaroo! oharaito icltara areni kimaclte iru!
+SG/OKA_0238.wav|2|seeshIkina meeshoowa wakaraNga, booni shiroi kamiga fusafUsato tsuite ite, kaNnushiga wasawasato furu yatsuda!
+SG/MAY_0088.wav|0|ahaha, imano okariNno setsumee, yutoriclpoinee.
+SG/MAY_0089.wav|0|Nja, mayu shiiwa korekara baitodakara, moo ikunee.
+SG/OKA_0239.wav|2|sooka. gaNbaclte koi. baitoga owacltara chokUsetsU kaeru noka?
+SG/MAY_0090.wav|0|uN.

filelists/nen_val.txt

@@ -0,0 +1,63 @@
+wav/nen001_001.wav|はい？呼びました？
+wav/nen001_002.wav|驚かせたならごめんなさい。通りかかったときにちょうど名前が聞こえてきたので
+wav/nen001_003.wav|私に何か用ですか？
+wav/nen001_004.wav|いえ、少し気になっただけですから、別に怒っているわけじゃないです。気にしないで下さい
+wav/nen001_005.wav|それより仮屋さん、例の件ですが――
+wav/nen001_006.wav|喜んでもらえたならなによりです
+wav/nen001_007.wav|また何かあったら、いつでも部室に来て下さい
+wav/nen001_008.wav|大したことじゃないので。それより体調の方はもういいんですか？
+wav/nen001_009.wav|それはよかったです。他に困ったことはありますか？
+wav/nen001_010.wav|何かあったらいつでも話して下さい。学院のことじゃなく、私事に関することでも何でも
+wav/nen001_011.wav|はい、いつでもどうぞ
+wav/nen001_012.wav|保科君も
+wav/nen001_013.wav|もし何か困ったことがあれば、力になりますから
+wav/nen001_014.wav|そうですか？私には、なにか悩み事があるように見えたりしましたけど……
+wav/nen001_015.wav|――なんて、言ってみただけですから、深い意味はありませんよ
+wav/nen001_016.wav|いえ、そういうわけじゃなくって……ただ何となく。そんな気がしただけですから
+wav/nen001_017.wav|あ、いえ、絶対に秘密というわけじゃないので気にしないで下さい
+wav/nen001_018.wav|好奇心だけで来られると困るので、本当に悩んでる人以外には、広めないようにお願いしているだけです
+wav/nen001_019.wav|そんな仰々しい話じゃなく……私は、オカ研に所属しているんです
+wav/nen001_020.wav|はい。そのオカ研です
+wav/nen001_021.wav|そうですよ。実は私たちが入学する前からあった部活なんです。今は部員がいなくて、所属してるのは私一人ですが
+wav/nen001_022.wav|なので最近は、学生会の方からは結構つつかれてて……活動も、発表などを意欲的にしているわけじゃありませんからね
+wav/nen001_023.wav|いえ、私は占いを。オカルトと言っても幅は広いので。それに部には私しかいませんから、結構好きにできるんです
+wav/nen001_024.wav|さすがに白蛇占いはできませんよ
+wav/nen001_025.wav|一応。知ってはいますけど、私にできるのはタロットぐらいです。あくまで趣味程度ですから
+wav/nen001_026.wav|あくまで占いの延長線上のものですから。人生相談なんていうほど大層な物じゃありません
+wav/nen001_027.wav|でも……もし保科君も嫌いでなければ、いつでも部室に来て下さい
+wav/nen001_028.wav|あ、はい。すぐに行きます
+wav/nen001_029.wav|それじゃあ私はこれで
+wav/nen001_030.wav|あの、先生
+wav/nen001_031.wav|はい。気付いたらこんな時間になってしまって
+wav/nen001_032.wav|図書室は……もう誰もいないんですか？
+wav/nen001_033.wav|その前にちょっと調べ物をさせて欲しいんですが、鍵を貸してもらえませんか？
+wav/nen001_034.wav|タロットカードのことで少々。時間はかかりません。５分……は無理かも……でも２０分もあれば終わりますから……お願いします
+wav/nen001_035.wav|わかりました。ありがとう……ございます
+wav/nen001_036.wav|は、はい………気をつけます……ハァ、ハァ……
+wav/nen001_037.wav|……ハァ……ハァ……
+wav/nen001_038.wav|……大丈夫、ですよね……んっ……
+wav/nen001_039.wav|……ハァ……ハァ……
+wav/nen001_040.wav|んっ、んん……
+wav/nen001_041.wav|あ……あっ、んっ……んんン、んッ、んッ、んぅぅッ……
+wav/nen001_042.wav|んっ、ふぅぅ……はぁ、はぁ、あっ、ああぁぁ……ふぁぁぁ……
+wav/nen001_043.wav|はぁ、はぁぁ……ん、ん、んっ……はぁぁ……ぁ、ぁ、ぁ……ぁぁ……ンッ……！
+wav/nen001_044.wav|あぁ、もう……こ、んなの、最低ですッ……はぁ、はぁ……学院内で、おっ、オナニー……をするなんてっ、ん、んんっ
+wav/nen001_045.wav|あっ、ああぁぁ……はぁ、はぁ、はぁぁぁ……ん、んンッッ
+wav/nen001_046.wav|はぁ、はぁ……私、なんてこと……あっ、あぁぁ……でも、気持ちよくて止まりません……ん、んんッ、ふぁ、あ、あぁぁぁ……
+wav/nen001_047.wav|うっ……あ、あ、あっ、あっ、ひあっ、ん、んはっ……はっ、はっ、んんぅぅッ
+wav/nen001_048.wav|はぁ……はぁ……んっ、んんん……んぁ、んぁ、あっ……んっ、ぃぃぃッ
+wav/nen001_049.wav|んっ、んっ、んくっ……ひっ、あっ、ぁっ、ぁっ、んんーーッ……
+wav/nen001_050.wav|はぁぁぁ……はぁ、はぁ……んんっ、んっ、んんん、んぁ……んぁ、ぁ、ぁ、ぁぁぁぁ……ッ！
+wav/nen001_051.wav|だめ、早く、しないと……先生が、戻って、きます……こんなところ、見られたら……んっ、んっ、んぁ、んぁ、あ、あ、あ、あ、あ……
+wav/nen001_052.wav|あ、あ、ん、んんんッ……はっ、はぁ、はぁ、あぁぁもう、本当に、最低ですっ……
+wav/nen001_053.wav|んッ……んはぁッ、はぁ、はぁ……ンン、んぁ、んぁぁ……あ、あ、あぁぁ……
+wav/nen001_054.wav|はぁー、はぁーーぁぁ、気持ちいい……あっ、あぃ、あぃっ、いい……本当に……んん
+wav/nen001_055.wav|早く……こんなこと、早く……先生が、また様子を……見に、きたりしない、内に……はぁ、はぁ、はぁ、はぁ
+wav/nen001_056.wav|はぁ、はぁ、はぁっ、はぁぁぁ……ぁ、ぁ、ぁ……ぁぁ……ぁぁんッ、んっ、んーッ
+wav/nen001_057.wav|んんッ、ん、んはぁぁぁーー……はぁ、はぁ、はぁァン、あッ、あん、あぁぁンッ
+wav/nen001_058.wav|はぁ、はぁ、はぁぁ……ヨダレ、でちゃう……じゅる……んぁっ、はぁ、はぁぁ……はっ、はっ、はっ……じゅる
+wav/nen001_059.wav|こんな、に、気持ちいいなんて、最低です……本当、図書室でオナニーなんて、最低すぎます……でも、でも
+wav/nen001_060.wav|あああぁぁ……今は、止められなくて……じゅる……はぁ、はぁぁ……あぁぁぁあぁ……
+wav/nen001_061.wav|はぁ、はぁ、はぁ……早く……あっ、あっ、あぁぁ……ぁぁぁ、早く、早く
+wav/nen001_062.wav|早く、イきたい、イきたい……このままイきたいぃ……はぁ、はぁ、はぁ、ぁぁ、ぁっ、ぁっ、あっ、あっ、あぁッ！
+wav/nen001_063.wav|あっ、あっ、ああぁぁ……痺れて、きたぁ……だめ、イきそう……んっ、だめじゃなくて、あっ、あっ、早く、このまま……早く早く早く

filelists/nen_val.txt.cleaned

@@ -0,0 +1,63 @@
+wav/nen001_001.wav|hai? yobimashIta?
+wav/nen001_002.wav|odorokasetanara gomeNnasai. toorikakaclta tokini choodo namaega kIkoete kitanode.
+wav/nen001_003.wav|watashini nanikayoodesUka?
+wav/nen001_004.wav|ie, sUkoshI kini nacltadakedesUkara, betsuni okoclte iru wakejanaidesU. kini shinaide kudasai.
+wav/nen001_005.wav|soreyori kariyasaN, reeno keNdesUga----
+wav/nen001_006.wav|yorokoNde moraetanara naniyoridesU.
+wav/nen001_007.wav|mata nanika acltara, itsudemo bushItsuni kite kudasai.
+wav/nen001_008.wav|taishIta kotojanainode. soreyori taichoono hoowa moo ii NdesUka?
+wav/nen001_009.wav|sorewa yokacltadesU. tani komaclta kotowa arimasUka?
+wav/nen001_010.wav|nanika acltara itsudemo hanashIte kudasai. gakuiNno kotojanaku, shijini kaNsuru kotodemo nanidemo.
+wav/nen001_011.wav|hai, itsudemo doozo.
+wav/nen001_012.wav|hoshinakuNmo.
+wav/nen001_013.wav|moshi nanika komaclta kotoga areba, chIkarani narimasUkara.
+wav/nen001_014.wav|soodesUka? watashiniwa, nanika nayamigotoga aru yooni mietari shimashItakedo......
+wav/nen001_015.wav|---- naNte, iclte mitadakedesUkara, fUkai imiwaarimaseNyo.
+wav/nen001_016.wav|ie, sooyuu wakejanakuclte...... tada naNtonaku. soNna kiga shItadakedesUkara.
+wav/nen001_017.wav|a, ie, zecltaini himitsUto iu wakejanainode kini shinaide kudasai.
+wav/nen001_018.wav|kookIshiNdakede korareruto komarunode, hoNtooni nayaNde runiNigainiwa, hiromenai yooni onegai shIte irudakedesU.
+wav/nen001_019.wav|soNna gyoogyooshii hanashijanaku...... watashiwa, okakeNni shozoku shIte iru NdesU.
+wav/nen001_020.wav|hai. sono okakeNdesU.
+wav/nen001_021.wav|soodesUyo. jitsuwa watashitachiga nyuugakU suru maekara aclta bukatsuna NdesU. imawa buiNga inakUte, shozoku shIteru nowa watashI hitoridesUga.
+wav/nen001_022.wav|nanode saikiNwa, gakUseekaino hookarawa keclkoo tsUtsukaretete...... katsudoomo, haclpyoonadoo iyokutekini shIte iru wakejaarimaseNkarane.
+wav/nen001_023.wav|ie, watashiwa uranaio. okarutoto icltemo habawa hiroinode. soreni buniwa watashIshika imaseNkara, keclkoo sUkini dekiru NdesU.
+wav/nen001_024.wav|sasugani shirohebiuranaiwa dekimaseNyo.
+wav/nen001_025.wav|ichioo. shiclte haimasUkedo, watashini dekiru nowa tarocltoguraidesU. akumade shumiteedodesUkara.
+wav/nen001_026.wav|akumade uranaino eNchooseNjoono monodesUkara. jiNseesoodaNnaNte iuhodo taisoona monojaarimaseN.
+wav/nen001_027.wav|demo...... moshI hoshinakuNmo kiraidenakereba, itsudemo bushItsuni kite kudasai.
+wav/nen001_028.wav|a, hai. suguni ikimasU.
+wav/nen001_029.wav|sorejaa watashiwa korede.
+wav/nen001_030.wav|ano, seNsee.
+wav/nen001_031.wav|hai. kizuitara koNna jikaNni naclte shimaclte.
+wav/nen001_032.wav|toshoshItsuwa...... moo daremo inai NdesUka?
+wav/nen001_033.wav|sono maeni choclto shirabebutsuo sasetehoshii NdesUga, kagio kashIte moraemaseNka?
+wav/nen001_034.wav|tarocltokaadono kotode shooshoo. jikaNwa kakarimaseN. gofuN...... w a murikamo...... demo nijuclpuNmo areba owarimasUkara...... onegai shimasU.
+wav/nen001_035.wav|wakarimashIta. arigatoo...... gozaimasU.
+wav/nen001_036.wav|w a, hai......... kio tsUkemasU...... haa, haa......
+wav/nen001_037.wav|...... haa...... haa......
+wav/nen001_038.wav|...... daijoobu, desUyone...... Ncl......
+wav/nen001_039.wav|...... haa...... haa......
+wav/nen001_040.wav|Ncl, NN......
+wav/nen001_041.wav|a...... acl, Ncl...... NN N, Ncl, Ncl, Nuucl......
+wav/nen001_042.wav|Ncl, fuuu...... haa, haa, acl, aaaa...... faaa......
+wav/nen001_043.wav|haa, haaa...... N, N, Ncl...... haaa...... a, a, a...... aa...... Ncl......!
+wav/nen001_044.wav|aa, moo...... k o, Nna n o, saiteede sucl...... haa, haa...... gakuiNnaide, ocl, onanii...... o surunaNtecl, N, NNcl.
+wav/nen001_045.wav|acl, aaaa...... haa, haa, haaaa...... N, NNclcl.
+wav/nen001_046.wav|haa, haa...... watashi, naNte koto...... acl, aaa...... demo, kimochiyokUte tomarimaseN...... N, NNcl, f a, a, aaaa......
+wav/nen001_047.wav|ucl...... a, a, acl, acl, hiacl, N, Nhacl...... hacl, hacl, NNuucl.
+wav/nen001_048.wav|haa...... haa...... Ncl, NNN...... Na, Na, acl...... Ncl, iiicl.
+wav/nen001_049.wav|Ncl, Ncl, N kucl...... hicl, acl, acl, acl, NNNNcl......
+wav/nen001_050.wav|haaaa...... haa, haa...... NNcl, Ncl, NNN, Na...... Na, a, a, aaaa...... cl!
+wav/nen001_051.wav|dame, hayaku, shinaito...... seNseega, modoclte, kimasU...... koNna tokoro, miraretara...... Ncl, Ncl, Na, Na, a, a, a, a, a......
+wav/nen001_052.wav|a, a, N, NNN cl...... hacl, haa, haa, aaa moo, hoNtooni, saiteede sucl......
+wav/nen001_053.wav|Ncl...... Nhaacl, haa, haa...... NN, Na, Naa...... a, a, aaa......
+wav/nen001_054.wav|haaa, haaaaaa, kimochiii...... acl, ai, aicl, ii...... hoNtooni...... NN.
+wav/nen001_055.wav|hayaku...... koNna koto, hayaku...... seNseega, mata yoosuo...... mini, kItarishinai, uchini...... haa, haa, haa, haa.
+wav/nen001_056.wav|haa, haa, haacl, haaaa...... a, a, a...... aa...... aaNcl, Ncl, NNcl.
+wav/nen001_057.wav|NNcl, N, Nhaaaaaa...... haa, haa, haaaN, acl, aN, aaaNcl.
+wav/nen001_058.wav|haa, haa, haaa...... yodare, dechau...... juru...... Nacl, haa, haaa...... hacl, hacl, hacl...... juru.
+wav/nen001_059.wav|koNna, n i, kimochiiinaNte, saiteedesU...... hoNtoo, toshoshItsude onaniinaNte, saitee sugimasU...... demo, demo.
+wav/nen001_060.wav|aaaaa...... imawa, tomerarenakUte...... juru...... haa, haaa...... aaaaaa......
+wav/nen001_061.wav|haa, haa, haa...... hayaku...... acl, acl, aaa...... aaa, hayaku, hayaku.
+wav/nen001_062.wav|hayaku, ikitai, ikitai...... kono mamaikitaii...... haa, haa, haa, aa, acl, acl, acl, acl, aacl!
+wav/nen001_063.wav|acl, acl, aaaa...... shibirete, kita a...... dame, ikIsou...... Ncl, damejanakUte, acl, acl, hayaku, kono mama...... hayakUhayakUhayaku.

inference.ipynb

@@ -0,0 +1,200 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "%matplotlib inline\n",
+    "import matplotlib.pyplot as plt\n",
+    "import IPython.display as ipd\n",
+    "\n",
+    "import os\n",
+    "import json\n",
+    "import math\n",
+    "import torch\n",
+    "from torch import nn\n",
+    "from torch.nn import functional as F\n",
+    "from torch.utils.data import DataLoader\n",
+    "\n",
+    "import commons\n",
+    "import utils\n",
+    "from data_utils import TextAudioLoader, TextAudioCollate, TextAudioSpeakerLoader, TextAudioSpeakerCollate\n",
+    "from models import SynthesizerTrn\n",
+    "from text.symbols import symbols\n",
+    "from text import text_to_sequence\n",
+    "\n",
+    "from scipy.io.wavfile import write\n",
+    "\n",
+    "\n",
+    "def get_text(text, hps):\n",
+    "    text_norm = text_to_sequence(text, hps.data.text_cleaners)\n",
+    "    if hps.data.add_blank:\n",
+    "        text_norm = commons.intersperse(text_norm, 0)\n",
+    "    text_norm = torch.LongTensor(text_norm)\n",
+    "    return text_norm"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## LJ Speech"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "hps = utils.get_hparams_from_file(\"./configs/ljs_base.json\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "net_g = SynthesizerTrn(\n",
+    "    len(symbols),\n",
+    "    hps.data.filter_length // 2 + 1,\n",
+    "    hps.train.segment_size // hps.data.hop_length,\n",
+    "    **hps.model).cuda()\n",
+    "_ = net_g.eval()\n",
+    "\n",
+    "_ = utils.load_checkpoint(\"/path/to/pretrained_ljs.pth\", net_g, None)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "stn_tst = get_text(\"VITS is Awesome!\", hps)\n",
+    "with torch.no_grad():\n",
+    "    x_tst = stn_tst.cuda().unsqueeze(0)\n",
+    "    x_tst_lengths = torch.LongTensor([stn_tst.size(0)]).cuda()\n",
+    "    audio = net_g.infer(x_tst, x_tst_lengths, noise_scale=.667, noise_scale_w=0.8, length_scale=1)[0][0,0].data.cpu().float().numpy()\n",
+    "ipd.display(ipd.Audio(audio, rate=hps.data.sampling_rate, normalize=False))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## VCTK"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "hps = utils.get_hparams_from_file(\"./configs/vctk_base.json\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "net_g = SynthesizerTrn(\n",
+    "    len(symbols),\n",
+    "    hps.data.filter_length // 2 + 1,\n",
+    "    hps.train.segment_size // hps.data.hop_length,\n",
+    "    n_speakers=hps.data.n_speakers,\n",
+    "    **hps.model).cuda()\n",
+    "_ = net_g.eval()\n",
+    "\n",
+    "_ = utils.load_checkpoint(\"/path/to/pretrained_vctk.pth\", net_g, None)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "stn_tst = get_text(\"VITS is Awesome!\", hps)\n",
+    "with torch.no_grad():\n",
+    "    x_tst = stn_tst.cuda().unsqueeze(0)\n",
+    "    x_tst_lengths = torch.LongTensor([stn_tst.size(0)]).cuda()\n",
+    "    sid = torch.LongTensor([4]).cuda()\n",
+    "    audio = net_g.infer(x_tst, x_tst_lengths, sid=sid, noise_scale=.667, noise_scale_w=0.8, length_scale=1)[0][0,0].data.cpu().float().numpy()\n",
+    "ipd.display(ipd.Audio(audio, rate=hps.data.sampling_rate, normalize=False))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Voice Conversion"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "dataset = TextAudioSpeakerLoader(hps.data.validation_files, hps.data)\n",
+    "collate_fn = TextAudioSpeakerCollate()\n",
+    "loader = DataLoader(dataset, num_workers=8, shuffle=False,\n",
+    "    batch_size=1, pin_memory=True,\n",
+    "    drop_last=True, collate_fn=collate_fn)\n",
+    "data_list = list(loader)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "with torch.no_grad():\n",
+    "    x, x_lengths, spec, spec_lengths, y, y_lengths, sid_src = [x.cuda() for x in data_list[0]]\n",
+    "    sid_tgt1 = torch.LongTensor([1]).cuda()\n",
+    "    sid_tgt2 = torch.LongTensor([2]).cuda()\n",
+    "    sid_tgt3 = torch.LongTensor([4]).cuda()\n",
+    "    audio1 = net_g.voice_conversion(spec, spec_lengths, sid_src=sid_src, sid_tgt=sid_tgt1)[0][0,0].data.cpu().float().numpy()\n",
+    "    audio2 = net_g.voice_conversion(spec, spec_lengths, sid_src=sid_src, sid_tgt=sid_tgt2)[0][0,0].data.cpu().float().numpy()\n",
+    "    audio3 = net_g.voice_conversion(spec, spec_lengths, sid_src=sid_src, sid_tgt=sid_tgt3)[0][0,0].data.cpu().float().numpy()\n",
+    "print(\"Original SID: %d\" % sid_src.item())\n",
+    "ipd.display(ipd.Audio(y[0].cpu().numpy(), rate=hps.data.sampling_rate, normalize=False))\n",
+    "print(\"Converted SID: %d\" % sid_tgt1.item())\n",
+    "ipd.display(ipd.Audio(audio1, rate=hps.data.sampling_rate, normalize=False))\n",
+    "print(\"Converted SID: %d\" % sid_tgt2.item())\n",
+    "ipd.display(ipd.Audio(audio2, rate=hps.data.sampling_rate, normalize=False))\n",
+    "print(\"Converted SID: %d\" % sid_tgt3.item())\n",
+    "ipd.display(ipd.Audio(audio3, rate=hps.data.sampling_rate, normalize=False))"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.7.7"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 4
+}

losses.py

@@ -0,0 +1,61 @@
+import torch 
+from torch.nn import functional as F
+
+import commons
+
+
+def feature_loss(fmap_r, fmap_g):
+  loss = 0
+  for dr, dg in zip(fmap_r, fmap_g):
+    for rl, gl in zip(dr, dg):
+      rl = rl.float().detach()
+      gl = gl.float()
+      loss += torch.mean(torch.abs(rl - gl))
+
+  return loss * 2 
+
+
+def discriminator_loss(disc_real_outputs, disc_generated_outputs):
+  loss = 0
+  r_losses = []
+  g_losses = []
+  for dr, dg in zip(disc_real_outputs, disc_generated_outputs):
+    dr = dr.float()
+    dg = dg.float()
+    r_loss = torch.mean((1-dr)**2)
+    g_loss = torch.mean(dg**2)
+    loss += (r_loss + g_loss)
+    r_losses.append(r_loss.item())
+    g_losses.append(g_loss.item())
+
+  return loss, r_losses, g_losses
+
+
+def generator_loss(disc_outputs):
+  loss = 0
+  gen_losses = []
+  for dg in disc_outputs:
+    dg = dg.float()
+    l = torch.mean((1-dg)**2)
+    gen_losses.append(l)
+    loss += l
+
+  return loss, gen_losses
+
+
+def kl_loss(z_p, logs_q, m_p, logs_p, z_mask):
+  """
+  z_p, logs_q: [b, h, t_t]
+  m_p, logs_p: [b, h, t_t]
+  """
+  z_p = z_p.float()
+  logs_q = logs_q.float()
+  m_p = m_p.float()
+  logs_p = logs_p.float()
+  z_mask = z_mask.float()
+
+  kl = logs_p - logs_q - 0.5
+  kl += 0.5 * ((z_p - m_p)**2) * torch.exp(-2. * logs_p)
+  kl = torch.sum(kl * z_mask)
+  l = kl / torch.sum(z_mask)
+  return l

mel_processing.py

@@ -0,0 +1,119 @@
+import math
+import os
+import random
+import torch
+from torch import nn
+import torch.nn.functional as F
+import torch.utils.data
+import numpy as np
+
+import logging
+
+numba_logger = logging.getLogger('numba')
+numba_logger.setLevel(logging.WARNING)
+import warnings
+warnings.filterwarnings('ignore')
+import librosa
+import librosa.util as librosa_util
+from librosa.util import normalize, pad_center, tiny
+from scipy.signal import get_window
+from scipy.io.wavfile import read
+from librosa.filters import mel as librosa_mel_fn
+
+MAX_WAV_VALUE = 32768.0
+
+
+def dynamic_range_compression_torch(x, C=1, clip_val=1e-5):
+    """
+    PARAMS
+    ------
+    C: compression factor
+    """
+    return torch.log(torch.clamp(x, min=clip_val) * C)
+
+
+def dynamic_range_decompression_torch(x, C=1):
+    """
+    PARAMS
+    ------
+    C: compression factor used to compress
+    """
+    return torch.exp(x) / C
+
+
+def spectral_normalize_torch(magnitudes):
+    output = dynamic_range_compression_torch(magnitudes)
+    return output
+
+
+def spectral_de_normalize_torch(magnitudes):
+    output = dynamic_range_decompression_torch(magnitudes)
+    return output
+
+
+mel_basis = {}
+hann_window = {}
+
+
+def spectrogram_torch(y, n_fft, sampling_rate, hop_size, win_size, center=False):
+    if torch.min(y) < -1.:
+        print('min value is ', torch.min(y))
+    if torch.max(y) > 1.:
+        print('max value is ', torch.max(y))
+
+    global hann_window
+    dtype_device = str(y.dtype) + '_' + str(y.device)
+    wnsize_dtype_device = str(win_size) + '_' + dtype_device
+    if wnsize_dtype_device not in hann_window:
+        hann_window[wnsize_dtype_device] = torch.hann_window(win_size).to(dtype=y.dtype, device=y.device)
+
+    y = torch.nn.functional.pad(y.unsqueeze(1), (int((n_fft-hop_size)/2), int((n_fft-hop_size)/2)), mode='reflect')
+    y = y.squeeze(1)
+
+    spec = torch.stft(y, n_fft, hop_length=hop_size, win_length=win_size, window=hann_window[wnsize_dtype_device],
+                      center=center, pad_mode='reflect', normalized=False, onesided=True)
+
+    spec = torch.sqrt(spec.pow(2).sum(-1) + 1e-6)
+    return spec
+
+
+def spec_to_mel_torch(spec, n_fft, num_mels, sampling_rate, fmin, fmax):
+    global mel_basis
+    dtype_device = str(spec.dtype) + '_' + str(spec.device)
+    fmax_dtype_device = str(fmax) + '_' + dtype_device
+    if fmax_dtype_device not in mel_basis:
+        mel = librosa_mel_fn(sampling_rate, n_fft, num_mels, fmin, fmax)
+        mel_basis[fmax_dtype_device] = torch.from_numpy(mel).to(dtype=spec.dtype, device=spec.device)
+    spec = torch.matmul(mel_basis[fmax_dtype_device], spec)
+    spec = spectral_normalize_torch(spec)
+    return spec
+
+
+def mel_spectrogram_torch(y, n_fft, num_mels, sampling_rate, hop_size, win_size, fmin, fmax, center=False):
+    if torch.min(y) < -1.:
+        print('min value is ', torch.min(y))
+    if torch.max(y) > 1.:
+        print('max value is ', torch.max(y))
+
+    global mel_basis, hann_window
+    dtype_device = str(y.dtype) + '_' + str(y.device)
+    fmax_dtype_device = str(fmax) + '_' + dtype_device
+    wnsize_dtype_device = str(win_size) + '_' + dtype_device
+    if fmax_dtype_device not in mel_basis:
+        mel = librosa_mel_fn(sampling_rate, n_fft, num_mels, fmin, fmax)
+        mel_basis[fmax_dtype_device] = torch.from_numpy(mel).to(dtype=y.dtype, device=y.device)
+    if wnsize_dtype_device not in hann_window:
+        hann_window[wnsize_dtype_device] = torch.hann_window(win_size).to(dtype=y.dtype, device=y.device)
+
+    y = torch.nn.functional.pad(y.unsqueeze(1), (int((n_fft-hop_size)/2), int((n_fft-hop_size)/2)), mode='reflect')
+    y = y.squeeze(1)
+
+    spec = torch.stft(y, n_fft, hop_length=hop_size, win_length=win_size, window=hann_window[wnsize_dtype_device],
+                      center=center, pad_mode='reflect', normalized=False, onesided=True)
+
+    spec = torch.sqrt(spec.pow(2).sum(-1) + 1e-6)
+
+    spec = torch.matmul(mel_basis[fmax_dtype_device], spec)
+    spec = spectral_normalize_torch(spec)
+
+    return spec

models.py

@@ -0,0 +1,534 @@
+import copy
+import math
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+import commons
+import modules
+import attentions
+import monotonic_align
+
+from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
+from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
+from commons import init_weights, get_padding
+
+
+class StochasticDurationPredictor(nn.Module):
+  def __init__(self, in_channels, filter_channels, kernel_size, p_dropout, n_flows=4, gin_channels=0):
+    super().__init__()
+    filter_channels = in_channels # it needs to be removed from future version.
+    self.in_channels = in_channels
+    self.filter_channels = filter_channels
+    self.kernel_size = kernel_size
+    self.p_dropout = p_dropout
+    self.n_flows = n_flows
+    self.gin_channels = gin_channels
+
+    self.log_flow = modules.Log()
+    self.flows = nn.ModuleList()
+    self.flows.append(modules.ElementwiseAffine(2))
+    for i in range(n_flows):
+      self.flows.append(modules.ConvFlow(2, filter_channels, kernel_size, n_layers=3))
+      self.flows.append(modules.Flip())
+
+    self.post_pre = nn.Conv1d(1, filter_channels, 1)
+    self.post_proj = nn.Conv1d(filter_channels, filter_channels, 1)
+    self.post_convs = modules.DDSConv(filter_channels, kernel_size, n_layers=3, p_dropout=p_dropout)
+    self.post_flows = nn.ModuleList()
+    self.post_flows.append(modules.ElementwiseAffine(2))
+    for i in range(4):
+      self.post_flows.append(modules.ConvFlow(2, filter_channels, kernel_size, n_layers=3))
+      self.post_flows.append(modules.Flip())
+
+    self.pre = nn.Conv1d(in_channels, filter_channels, 1)
+    self.proj = nn.Conv1d(filter_channels, filter_channels, 1)
+    self.convs = modules.DDSConv(filter_channels, kernel_size, n_layers=3, p_dropout=p_dropout)
+    if gin_channels != 0:
+      self.cond = nn.Conv1d(gin_channels, filter_channels, 1)
+
+  def forward(self, x, x_mask, w=None, g=None, reverse=False, noise_scale=1.0):
+    x = torch.detach(x)
+    x = self.pre(x)
+    if g is not None:
+      g = torch.detach(g)
+      x = x + self.cond(g)
+    x = self.convs(x, x_mask)
+    x = self.proj(x) * x_mask
+
+    if not reverse:
+      flows = self.flows
+      assert w is not None
+
+      logdet_tot_q = 0 
+      h_w = self.post_pre(w)
+      h_w = self.post_convs(h_w, x_mask)
+      h_w = self.post_proj(h_w) * x_mask
+      e_q = torch.randn(w.size(0), 2, w.size(2)).to(device=x.device, dtype=x.dtype) * x_mask
+      z_q = e_q
+      for flow in self.post_flows:
+        z_q, logdet_q = flow(z_q, x_mask, g=(x + h_w))
+        logdet_tot_q += logdet_q
+      z_u, z1 = torch.split(z_q, [1, 1], 1) 
+      u = torch.sigmoid(z_u) * x_mask
+      z0 = (w - u) * x_mask
+      logdet_tot_q += torch.sum((F.logsigmoid(z_u) + F.logsigmoid(-z_u)) * x_mask, [1,2])
+      logq = torch.sum(-0.5 * (math.log(2*math.pi) + (e_q**2)) * x_mask, [1,2]) - logdet_tot_q
+
+      logdet_tot = 0
+      z0, logdet = self.log_flow(z0, x_mask)
+      logdet_tot += logdet
+      z = torch.cat([z0, z1], 1)
+      for flow in flows:
+        z, logdet = flow(z, x_mask, g=x, reverse=reverse)
+        logdet_tot = logdet_tot + logdet
+      nll = torch.sum(0.5 * (math.log(2*math.pi) + (z**2)) * x_mask, [1,2]) - logdet_tot
+      return nll + logq # [b]
+    else:
+      flows = list(reversed(self.flows))
+      flows = flows[:-2] + [flows[-1]] # remove a useless vflow
+      z = torch.randn(x.size(0), 2, x.size(2)).to(device=x.device, dtype=x.dtype) * noise_scale
+      for flow in flows:
+        z = flow(z, x_mask, g=x, reverse=reverse)
+      z0, z1 = torch.split(z, [1, 1], 1)
+      logw = z0
+      return logw
+
+
+class DurationPredictor(nn.Module):
+  def __init__(self, in_channels, filter_channels, kernel_size, p_dropout, gin_channels=0):
+    super().__init__()
+
+    self.in_channels = in_channels
+    self.filter_channels = filter_channels
+    self.kernel_size = kernel_size
+    self.p_dropout = p_dropout
+    self.gin_channels = gin_channels
+
+    self.drop = nn.Dropout(p_dropout)
+    self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size, padding=kernel_size//2)
+    self.norm_1 = modules.LayerNorm(filter_channels)
+    self.conv_2 = nn.Conv1d(filter_channels, filter_channels, kernel_size, padding=kernel_size//2)
+    self.norm_2 = modules.LayerNorm(filter_channels)
+    self.proj = nn.Conv1d(filter_channels, 1, 1)
+
+    if gin_channels != 0:
+      self.cond = nn.Conv1d(gin_channels, in_channels, 1)
+
+  def forward(self, x, x_mask, g=None):
+    x = torch.detach(x)
+    if g is not None:
+      g = torch.detach(g)
+      x = x + self.cond(g)
+    x = self.conv_1(x * x_mask)
+    x = torch.relu(x)
+    x = self.norm_1(x)
+    x = self.drop(x)
+    x = self.conv_2(x * x_mask)
+    x = torch.relu(x)
+    x = self.norm_2(x)
+    x = self.drop(x)
+    x = self.proj(x * x_mask)
+    return x * x_mask
+
+
+class TextEncoder(nn.Module):
+  def __init__(self,
+      n_vocab,
+      out_channels,
+      hidden_channels,
+      filter_channels,
+      n_heads,
+      n_layers,
+      kernel_size,
+      p_dropout):
+    super().__init__()
+    self.n_vocab = n_vocab
+    self.out_channels = out_channels
+    self.hidden_channels = hidden_channels
+    self.filter_channels = filter_channels
+    self.n_heads = n_heads
+    self.n_layers = n_layers
+    self.kernel_size = kernel_size
+    self.p_dropout = p_dropout
+
+    self.emb = nn.Embedding(n_vocab, hidden_channels)
+    nn.init.normal_(self.emb.weight, 0.0, hidden_channels**-0.5)
+
+    self.encoder = attentions.Encoder(
+      hidden_channels,
+      filter_channels,
+      n_heads,
+      n_layers,
+      kernel_size,
+      p_dropout)
+    self.proj= nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+  def forward(self, x, x_lengths):
+    x = self.emb(x) * math.sqrt(self.hidden_channels) # [b, t, h]
+    x = torch.transpose(x, 1, -1) # [b, h, t]
+    x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(x.dtype)
+
+    x = self.encoder(x * x_mask, x_mask)
+    stats = self.proj(x) * x_mask
+
+    m, logs = torch.split(stats, self.out_channels, dim=1)
+    return x, m, logs, x_mask
+
+
+class ResidualCouplingBlock(nn.Module):
+  def __init__(self,
+      channels,
+      hidden_channels,
+      kernel_size,
+      dilation_rate,
+      n_layers,
+      n_flows=4,
+      gin_channels=0):
+    super().__init__()
+    self.channels = channels
+    self.hidden_channels = hidden_channels
+    self.kernel_size = kernel_size
+    self.dilation_rate = dilation_rate
+    self.n_layers = n_layers
+    self.n_flows = n_flows
+    self.gin_channels = gin_channels
+
+    self.flows = nn.ModuleList()
+    for i in range(n_flows):
+      self.flows.append(modules.ResidualCouplingLayer(channels, hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=gin_channels, mean_only=True))
+      self.flows.append(modules.Flip())
+
+  def forward(self, x, x_mask, g=None, reverse=False):
+    if not reverse:
+      for flow in self.flows:
+        x, _ = flow(x, x_mask, g=g, reverse=reverse)
+    else:
+      for flow in reversed(self.flows):
+        x = flow(x, x_mask, g=g, reverse=reverse)
+    return x
+
+
+class PosteriorEncoder(nn.Module):
+  def __init__(self,
+      in_channels,
+      out_channels,
+      hidden_channels,
+      kernel_size,
+      dilation_rate,
+      n_layers,
+      gin_channels=0):
+    super().__init__()
+    self.in_channels = in_channels
+    self.out_channels = out_channels
+    self.hidden_channels = hidden_channels
+    self.kernel_size = kernel_size
+    self.dilation_rate = dilation_rate
+    self.n_layers = n_layers
+    self.gin_channels = gin_channels
+
+    self.pre = nn.Conv1d(in_channels, hidden_channels, 1)
+    self.enc = modules.WN(hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=gin_channels)
+    self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+  def forward(self, x, x_lengths, g=None):
+    x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(x.dtype)
+    x = self.pre(x) * x_mask
+    x = self.enc(x, x_mask, g=g)
+    stats = self.proj(x) * x_mask
+    m, logs = torch.split(stats, self.out_channels, dim=1)
+    z = (m + torch.randn_like(m) * torch.exp(logs)) * x_mask
+    return z, m, logs, x_mask
+
+
+class Generator(torch.nn.Module):
+    def __init__(self, initial_channel, resblock, resblock_kernel_sizes, resblock_dilation_sizes, upsample_rates, upsample_initial_channel, upsample_kernel_sizes, gin_channels=0):
+        super(Generator, self).__init__()
+        self.num_kernels = len(resblock_kernel_sizes)
+        self.num_upsamples = len(upsample_rates)
+        self.conv_pre = Conv1d(initial_channel, upsample_initial_channel, 7, 1, padding=3)
+        resblock = modules.ResBlock1 if resblock == '1' else modules.ResBlock2
+
+        self.ups = nn.ModuleList()
+        for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
+            self.ups.append(weight_norm(
+                ConvTranspose1d(upsample_initial_channel//(2**i), upsample_initial_channel//(2**(i+1)),
+                                k, u, padding=(k-u)//2)))
+
+        self.resblocks = nn.ModuleList()
+        for i in range(len(self.ups)):
+            ch = upsample_initial_channel//(2**(i+1))
+            for j, (k, d) in enumerate(zip(resblock_kernel_sizes, resblock_dilation_sizes)):
+                self.resblocks.append(resblock(ch, k, d))
+
+        self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
+        self.ups.apply(init_weights)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
+
+    def forward(self, x, g=None):
+        x = self.conv_pre(x)
+        if g is not None:
+          x = x + self.cond(g)
+
+        for i in range(self.num_upsamples):
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            x = self.ups[i](x)
+            xs = None
+            for j in range(self.num_kernels):
+                if xs is None:
+                    xs = self.resblocks[i*self.num_kernels+j](x)
+                else:
+                    xs += self.resblocks[i*self.num_kernels+j](x)
+            x = xs / self.num_kernels
+        x = F.leaky_relu(x)
+        x = self.conv_post(x)
+        x = torch.tanh(x)
+
+        return x
+
+    def remove_weight_norm(self):
+        print('Removing weight norm...')
+        for l in self.ups:
+            remove_weight_norm(l)
+        for l in self.resblocks:
+            l.remove_weight_norm()
+
+
+class DiscriminatorP(torch.nn.Module):
+    def __init__(self, period, kernel_size=5, stride=3, use_spectral_norm=False):
+        super(DiscriminatorP, self).__init__()
+        self.period = period
+        self.use_spectral_norm = use_spectral_norm
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList([
+            norm_f(Conv2d(1, 32, (kernel_size, 1), (stride, 1), padding=(get_padding(kernel_size, 1), 0))),
+            norm_f(Conv2d(32, 128, (kernel_size, 1), (stride, 1), padding=(get_padding(kernel_size, 1), 0))),
+            norm_f(Conv2d(128, 512, (kernel_size, 1), (stride, 1), padding=(get_padding(kernel_size, 1), 0))),
+            norm_f(Conv2d(512, 1024, (kernel_size, 1), (stride, 1), padding=(get_padding(kernel_size, 1), 0))),
+            norm_f(Conv2d(1024, 1024, (kernel_size, 1), 1, padding=(get_padding(kernel_size, 1), 0))),
+        ])
+        self.conv_post = norm_f(Conv2d(1024, 1, (3, 1), 1, padding=(1, 0)))
+
+    def forward(self, x):
+        fmap = []
+
+        # 1d to 2d
+        b, c, t = x.shape
+        if t % self.period != 0: # pad first
+            n_pad = self.period - (t % self.period)
+            x = F.pad(x, (0, n_pad), "reflect")
+            t = t + n_pad
+        x = x.view(b, c, t // self.period, self.period)
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap
+
+
+class DiscriminatorS(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(DiscriminatorS, self).__init__()
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList([
+            norm_f(Conv1d(1, 16, 15, 1, padding=7)),
+            norm_f(Conv1d(16, 64, 41, 4, groups=4, padding=20)),
+            norm_f(Conv1d(64, 256, 41, 4, groups=16, padding=20)),
+            norm_f(Conv1d(256, 1024, 41, 4, groups=64, padding=20)),
+            norm_f(Conv1d(1024, 1024, 41, 4, groups=256, padding=20)),
+            norm_f(Conv1d(1024, 1024, 5, 1, padding=2)),
+        ])
+        self.conv_post = norm_f(Conv1d(1024, 1, 3, 1, padding=1))
+
+    def forward(self, x):
+        fmap = []
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap
+
+
+class MultiPeriodDiscriminator(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminator, self).__init__()
+        periods = [2,3,5,7,11]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
+
+class SynthesizerTrn(nn.Module):
+  """
+  Synthesizer for Training
+  """
+
+  def __init__(self, 
+    n_vocab,
+    spec_channels,
+    segment_size,
+    inter_channels,
+    hidden_channels,
+    filter_channels,
+    n_heads,
+    n_layers,
+    kernel_size,
+    p_dropout,
+    resblock, 
+    resblock_kernel_sizes, 
+    resblock_dilation_sizes, 
+    upsample_rates, 
+    upsample_initial_channel, 
+    upsample_kernel_sizes,
+    n_speakers=0,
+    gin_channels=0,
+    use_sdp=True,
+    **kwargs):
+
+    super().__init__()
+    self.n_vocab = n_vocab
+    self.spec_channels = spec_channels
+    self.inter_channels = inter_channels
+    self.hidden_channels = hidden_channels
+    self.filter_channels = filter_channels
+    self.n_heads = n_heads
+    self.n_layers = n_layers
+    self.kernel_size = kernel_size
+    self.p_dropout = p_dropout
+    self.resblock = resblock
+    self.resblock_kernel_sizes = resblock_kernel_sizes
+    self.resblock_dilation_sizes = resblock_dilation_sizes
+    self.upsample_rates = upsample_rates
+    self.upsample_initial_channel = upsample_initial_channel
+    self.upsample_kernel_sizes = upsample_kernel_sizes
+    self.segment_size = segment_size
+    self.n_speakers = n_speakers
+    self.gin_channels = gin_channels
+
+    self.use_sdp = use_sdp
+
+    self.enc_p = TextEncoder(n_vocab,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout)
+    self.dec = Generator(inter_channels, resblock, resblock_kernel_sizes, resblock_dilation_sizes, upsample_rates, upsample_initial_channel, upsample_kernel_sizes, gin_channels=gin_channels)
+    self.enc_q = PosteriorEncoder(spec_channels, inter_channels, hidden_channels, 5, 1, 16, gin_channels=gin_channels)
+    self.flow = ResidualCouplingBlock(inter_channels, hidden_channels, 5, 1, 4, gin_channels=gin_channels)
+
+    if use_sdp:
+      self.dp = StochasticDurationPredictor(hidden_channels, 192, 3, 0.5, 4, gin_channels=gin_channels)
+    else:
+      self.dp = DurationPredictor(hidden_channels, 256, 3, 0.5, gin_channels=gin_channels)
+
+    if n_speakers > 1:
+      self.emb_g = nn.Embedding(n_speakers, gin_channels)
+
+  def forward(self, x, x_lengths, y, y_lengths, sid=None):
+
+    x, m_p, logs_p, x_mask = self.enc_p(x, x_lengths)
+    if self.n_speakers > 0:
+      g = self.emb_g(sid).unsqueeze(-1) # [b, h, 1]
+    else:
+      g = None
+
+    z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
+    z_p = self.flow(z, y_mask, g=g)
+
+    with torch.no_grad():
+      # negative cross-entropy
+      s_p_sq_r = torch.exp(-2 * logs_p) # [b, d, t]
+      neg_cent1 = torch.sum(-0.5 * math.log(2 * math.pi) - logs_p, [1], keepdim=True) # [b, 1, t_s]
+      neg_cent2 = torch.matmul(-0.5 * (z_p ** 2).transpose(1, 2), s_p_sq_r) # [b, t_t, d] x [b, d, t_s] = [b, t_t, t_s]
+      neg_cent3 = torch.matmul(z_p.transpose(1, 2), (m_p * s_p_sq_r)) # [b, t_t, d] x [b, d, t_s] = [b, t_t, t_s]
+      neg_cent4 = torch.sum(-0.5 * (m_p ** 2) * s_p_sq_r, [1], keepdim=True) # [b, 1, t_s]
+      neg_cent = neg_cent1 + neg_cent2 + neg_cent3 + neg_cent4
+
+      attn_mask = torch.unsqueeze(x_mask, 2) * torch.unsqueeze(y_mask, -1)
+      attn = monotonic_align.maximum_path(neg_cent, attn_mask.squeeze(1)).unsqueeze(1).detach()
+
+    w = attn.sum(2)
+    if self.use_sdp:
+      l_length = self.dp(x, x_mask, w, g=g)
+      l_length = l_length / torch.sum(x_mask)
+    else:
+      logw_ = torch.log(w + 1e-6) * x_mask
+      logw = self.dp(x, x_mask, g=g)
+      l_length = torch.sum((logw - logw_)**2, [1,2]) / torch.sum(x_mask) # for averaging 
+
+    # expand prior
+    m_p = torch.matmul(attn.squeeze(1), m_p.transpose(1, 2)).transpose(1, 2)
+    logs_p = torch.matmul(attn.squeeze(1), logs_p.transpose(1, 2)).transpose(1, 2)
+
+    z_slice, ids_slice = commons.rand_slice_segments(z, y_lengths, self.segment_size)
+    o = self.dec(z_slice, g=g)
+    return o, l_length, attn, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
+
+  def infer(self, x, x_lengths, sid=None, noise_scale=1, length_scale=1, noise_scale_w=1., max_len=None):
+    x, m_p, logs_p, x_mask = self.enc_p(x, x_lengths)
+    if self.n_speakers > 0:
+      g = self.emb_g(sid).unsqueeze(-1) # [b, h, 1]
+    else:
+      g = None
+
+    if self.use_sdp:
+      logw = self.dp(x, x_mask, g=g, reverse=True, noise_scale=noise_scale_w)
+    else:
+      logw = self.dp(x, x_mask, g=g)
+    w = torch.exp(logw) * x_mask * length_scale
+    w_ceil = torch.ceil(w)
+    y_lengths = torch.clamp_min(torch.sum(w_ceil, [1, 2]), 1).long()
+    y_mask = torch.unsqueeze(commons.sequence_mask(y_lengths, None), 1).to(x_mask.dtype)
+    attn_mask = torch.unsqueeze(x_mask, 2) * torch.unsqueeze(y_mask, -1)
+    attn = commons.generate_path(w_ceil, attn_mask)
+
+    m_p = torch.matmul(attn.squeeze(1), m_p.transpose(1, 2)).transpose(1, 2) # [b, t', t], [b, t, d] -> [b, d, t']
+    logs_p = torch.matmul(attn.squeeze(1), logs_p.transpose(1, 2)).transpose(1, 2) # [b, t', t], [b, t, d] -> [b, d, t']
+
+    z_p = m_p + torch.randn_like(m_p) * torch.exp(logs_p) * noise_scale
+    z = self.flow(z_p, y_mask, g=g, reverse=True)
+    o = self.dec((z * y_mask)[:,:,:max_len], g=g)
+    return o, attn, y_mask, (z, z_p, m_p, logs_p)
+
+  def voice_conversion(self, y, y_lengths, sid_src, sid_tgt):
+    assert self.n_speakers > 0, "n_speakers have to be larger than 0."
+    g_src = self.emb_g(sid_src).unsqueeze(-1)
+    g_tgt = self.emb_g(sid_tgt).unsqueeze(-1)
+    z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g_src)
+    z_p = self.flow(z, y_mask, g=g_src)
+    z_hat = self.flow(z_p, y_mask, g=g_tgt, reverse=True)
+    o_hat = self.dec(z_hat * y_mask, g=g_tgt)
+    return o_hat, y_mask, (z, z_p, z_hat)
+

modules.py

@@ -0,0 +1,390 @@
+import copy
+import math
+import numpy as np
+import scipy
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
+from torch.nn.utils import weight_norm, remove_weight_norm
+
+import commons
+from commons import init_weights, get_padding
+from transforms import piecewise_rational_quadratic_transform
+
+
+LRELU_SLOPE = 0.1
+
+
+class LayerNorm(nn.Module):
+  def __init__(self, channels, eps=1e-5):
+    super().__init__()
+    self.channels = channels
+    self.eps = eps
+
+    self.gamma = nn.Parameter(torch.ones(channels))
+    self.beta = nn.Parameter(torch.zeros(channels))
+
+  def forward(self, x):
+    x = x.transpose(1, -1)
+    x = F.layer_norm(x, (self.channels,), self.gamma, self.beta, self.eps)
+    return x.transpose(1, -1)
+
+ 
+class ConvReluNorm(nn.Module):
+  def __init__(self, in_channels, hidden_channels, out_channels, kernel_size, n_layers, p_dropout):
+    super().__init__()
+    self.in_channels = in_channels
+    self.hidden_channels = hidden_channels
+    self.out_channels = out_channels
+    self.kernel_size = kernel_size
+    self.n_layers = n_layers
+    self.p_dropout = p_dropout
+    assert n_layers > 1, "Number of layers should be larger than 0."
+
+    self.conv_layers = nn.ModuleList()
+    self.norm_layers = nn.ModuleList()
+    self.conv_layers.append(nn.Conv1d(in_channels, hidden_channels, kernel_size, padding=kernel_size//2))
+    self.norm_layers.append(LayerNorm(hidden_channels))
+    self.relu_drop = nn.Sequential(
+        nn.ReLU(),
+        nn.Dropout(p_dropout))
+    for _ in range(n_layers-1):
+      self.conv_layers.append(nn.Conv1d(hidden_channels, hidden_channels, kernel_size, padding=kernel_size//2))
+      self.norm_layers.append(LayerNorm(hidden_channels))
+    self.proj = nn.Conv1d(hidden_channels, out_channels, 1)
+    self.proj.weight.data.zero_()
+    self.proj.bias.data.zero_()
+
+  def forward(self, x, x_mask):
+    x_org = x
+    for i in range(self.n_layers):
+      x = self.conv_layers[i](x * x_mask)
+      x = self.norm_layers[i](x)
+      x = self.relu_drop(x)
+    x = x_org + self.proj(x)
+    return x * x_mask
+
+
+class DDSConv(nn.Module):
+  """
+  Dialted and Depth-Separable Convolution
+  """
+  def __init__(self, channels, kernel_size, n_layers, p_dropout=0.):
+    super().__init__()
+    self.channels = channels
+    self.kernel_size = kernel_size
+    self.n_layers = n_layers
+    self.p_dropout = p_dropout
+
+    self.drop = nn.Dropout(p_dropout)
+    self.convs_sep = nn.ModuleList()
+    self.convs_1x1 = nn.ModuleList()
+    self.norms_1 = nn.ModuleList()
+    self.norms_2 = nn.ModuleList()
+    for i in range(n_layers):
+      dilation = kernel_size ** i
+      padding = (kernel_size * dilation - dilation) // 2
+      self.convs_sep.append(nn.Conv1d(channels, channels, kernel_size, 
+          groups=channels, dilation=dilation, padding=padding
+      ))
+      self.convs_1x1.append(nn.Conv1d(channels, channels, 1))
+      self.norms_1.append(LayerNorm(channels))
+      self.norms_2.append(LayerNorm(channels))
+
+  def forward(self, x, x_mask, g=None):
+    if g is not None:
+      x = x + g
+    for i in range(self.n_layers):
+      y = self.convs_sep[i](x * x_mask)
+      y = self.norms_1[i](y)
+      y = F.gelu(y)
+      y = self.convs_1x1[i](y)
+      y = self.norms_2[i](y)
+      y = F.gelu(y)
+      y = self.drop(y)
+      x = x + y
+    return x * x_mask
+
+
+class WN(torch.nn.Module):
+  def __init__(self, hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=0, p_dropout=0):
+    super(WN, self).__init__()
+    assert(kernel_size % 2 == 1)
+    self.hidden_channels =hidden_channels
+    self.kernel_size = kernel_size,
+    self.dilation_rate = dilation_rate
+    self.n_layers = n_layers
+    self.gin_channels = gin_channels
+    self.p_dropout = p_dropout
+
+    self.in_layers = torch.nn.ModuleList()
+    self.res_skip_layers = torch.nn.ModuleList()
+    self.drop = nn.Dropout(p_dropout)
+
+    if gin_channels != 0:
+      cond_layer = torch.nn.Conv1d(gin_channels, 2*hidden_channels*n_layers, 1)
+      self.cond_layer = torch.nn.utils.weight_norm(cond_layer, name='weight')
+
+    for i in range(n_layers):
+      dilation = dilation_rate ** i
+      padding = int((kernel_size * dilation - dilation) / 2)
+      in_layer = torch.nn.Conv1d(hidden_channels, 2*hidden_channels, kernel_size,
+                                 dilation=dilation, padding=padding)
+      in_layer = torch.nn.utils.weight_norm(in_layer, name='weight')
+      self.in_layers.append(in_layer)
+
+      # last one is not necessary
+      if i < n_layers - 1:
+        res_skip_channels = 2 * hidden_channels
+      else:
+        res_skip_channels = hidden_channels
+
+      res_skip_layer = torch.nn.Conv1d(hidden_channels, res_skip_channels, 1)
+      res_skip_layer = torch.nn.utils.weight_norm(res_skip_layer, name='weight')
+      self.res_skip_layers.append(res_skip_layer)
+
+  def forward(self, x, x_mask, g=None, **kwargs):
+    output = torch.zeros_like(x)
+    n_channels_tensor = torch.IntTensor([self.hidden_channels])
+
+    if g is not None:
+      g = self.cond_layer(g)
+
+    for i in range(self.n_layers):
+      x_in = self.in_layers[i](x)
+      if g is not None:
+        cond_offset = i * 2 * self.hidden_channels
+        g_l = g[:,cond_offset:cond_offset+2*self.hidden_channels,:]
+      else:
+        g_l = torch.zeros_like(x_in)
+
+      acts = commons.fused_add_tanh_sigmoid_multiply(
+          x_in,
+          g_l,
+          n_channels_tensor)
+      acts = self.drop(acts)
+
+      res_skip_acts = self.res_skip_layers[i](acts)
+      if i < self.n_layers - 1:
+        res_acts = res_skip_acts[:,:self.hidden_channels,:]
+        x = (x + res_acts) * x_mask
+        output = output + res_skip_acts[:,self.hidden_channels:,:]
+      else:
+        output = output + res_skip_acts
+    return output * x_mask
+
+  def remove_weight_norm(self):
+    if self.gin_channels != 0:
+      torch.nn.utils.remove_weight_norm(self.cond_layer)
+    for l in self.in_layers:
+      torch.nn.utils.remove_weight_norm(l)
+    for l in self.res_skip_layers:
+     torch.nn.utils.remove_weight_norm(l)
+
+
+class ResBlock1(torch.nn.Module):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3, 5)):
+        super(ResBlock1, self).__init__()
+        self.convs1 = nn.ModuleList([
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[0],
+                               padding=get_padding(kernel_size, dilation[0]))),
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[1],
+                               padding=get_padding(kernel_size, dilation[1]))),
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[2],
+                               padding=get_padding(kernel_size, dilation[2])))
+        ])
+        self.convs1.apply(init_weights)
+
+        self.convs2 = nn.ModuleList([
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=1,
+                               padding=get_padding(kernel_size, 1))),
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=1,
+                               padding=get_padding(kernel_size, 1))),
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=1,
+                               padding=get_padding(kernel_size, 1)))
+        ])
+        self.convs2.apply(init_weights)
+
+    def forward(self, x, x_mask=None):
+        for c1, c2 in zip(self.convs1, self.convs2):
+            xt = F.leaky_relu(x, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c1(xt)
+            xt = F.leaky_relu(xt, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c2(xt)
+            x = xt + x
+        if x_mask is not None:
+            x = x * x_mask
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs1:
+            remove_weight_norm(l)
+        for l in self.convs2:
+            remove_weight_norm(l)
+
+
+class ResBlock2(torch.nn.Module):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3)):
+        super(ResBlock2, self).__init__()
+        self.convs = nn.ModuleList([
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[0],
+                               padding=get_padding(kernel_size, dilation[0]))),
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[1],
+                               padding=get_padding(kernel_size, dilation[1])))
+        ])
+        self.convs.apply(init_weights)
+
+    def forward(self, x, x_mask=None):
+        for c in self.convs:
+            xt = F.leaky_relu(x, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c(xt)
+            x = xt + x
+        if x_mask is not None:
+            x = x * x_mask
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs:
+            remove_weight_norm(l)
+
+
+class Log(nn.Module):
+  def forward(self, x, x_mask, reverse=False, **kwargs):
+    if not reverse:
+      y = torch.log(torch.clamp_min(x, 1e-5)) * x_mask
+      logdet = torch.sum(-y, [1, 2])
+      return y, logdet
+    else:
+      x = torch.exp(x) * x_mask
+      return x
+    
+
+class Flip(nn.Module):
+  def forward(self, x, *args, reverse=False, **kwargs):
+    x = torch.flip(x, [1])
+    if not reverse:
+      logdet = torch.zeros(x.size(0)).to(dtype=x.dtype, device=x.device)
+      return x, logdet
+    else:
+      return x
+
+
+class ElementwiseAffine(nn.Module):
+  def __init__(self, channels):
+    super().__init__()
+    self.channels = channels
+    self.m = nn.Parameter(torch.zeros(channels,1))
+    self.logs = nn.Parameter(torch.zeros(channels,1))
+
+  def forward(self, x, x_mask, reverse=False, **kwargs):
+    if not reverse:
+      y = self.m + torch.exp(self.logs) * x
+      y = y * x_mask
+      logdet = torch.sum(self.logs * x_mask, [1,2])
+      return y, logdet
+    else:
+      x = (x - self.m) * torch.exp(-self.logs) * x_mask
+      return x
+
+
+class ResidualCouplingLayer(nn.Module):
+  def __init__(self,
+      channels,
+      hidden_channels,
+      kernel_size,
+      dilation_rate,
+      n_layers,
+      p_dropout=0,
+      gin_channels=0,
+      mean_only=False):
+    assert channels % 2 == 0, "channels should be divisible by 2"
+    super().__init__()
+    self.channels = channels
+    self.hidden_channels = hidden_channels
+    self.kernel_size = kernel_size
+    self.dilation_rate = dilation_rate
+    self.n_layers = n_layers
+    self.half_channels = channels // 2
+    self.mean_only = mean_only
+
+    self.pre = nn.Conv1d(self.half_channels, hidden_channels, 1)
+    self.enc = WN(hidden_channels, kernel_size, dilation_rate, n_layers, p_dropout=p_dropout, gin_channels=gin_channels)
+    self.post = nn.Conv1d(hidden_channels, self.half_channels * (2 - mean_only), 1)
+    self.post.weight.data.zero_()
+    self.post.bias.data.zero_()
+
+  def forward(self, x, x_mask, g=None, reverse=False):
+    x0, x1 = torch.split(x, [self.half_channels]*2, 1)
+    h = self.pre(x0) * x_mask
+    h = self.enc(h, x_mask, g=g)
+    stats = self.post(h) * x_mask
+    if not self.mean_only:
+      m, logs = torch.split(stats, [self.half_channels]*2, 1)
+    else:
+      m = stats
+      logs = torch.zeros_like(m)
+
+    if not reverse:
+      x1 = m + x1 * torch.exp(logs) * x_mask
+      x = torch.cat([x0, x1], 1)
+      logdet = torch.sum(logs, [1,2])
+      return x, logdet
+    else:
+      x1 = (x1 - m) * torch.exp(-logs) * x_mask
+      x = torch.cat([x0, x1], 1)
+      return x
+
+
+class ConvFlow(nn.Module):
+  def __init__(self, in_channels, filter_channels, kernel_size, n_layers, num_bins=10, tail_bound=5.0):
+    super().__init__()
+    self.in_channels = in_channels
+    self.filter_channels = filter_channels
+    self.kernel_size = kernel_size
+    self.n_layers = n_layers
+    self.num_bins = num_bins
+    self.tail_bound = tail_bound
+    self.half_channels = in_channels // 2
+
+    self.pre = nn.Conv1d(self.half_channels, filter_channels, 1)
+    self.convs = DDSConv(filter_channels, kernel_size, n_layers, p_dropout=0.)
+    self.proj = nn.Conv1d(filter_channels, self.half_channels * (num_bins * 3 - 1), 1)
+    self.proj.weight.data.zero_()
+    self.proj.bias.data.zero_()
+
+  def forward(self, x, x_mask, g=None, reverse=False):
+    x0, x1 = torch.split(x, [self.half_channels]*2, 1)
+    h = self.pre(x0)
+    h = self.convs(h, x_mask, g=g)
+    h = self.proj(h) * x_mask
+
+    b, c, t = x0.shape
+    h = h.reshape(b, c, -1, t).permute(0, 1, 3, 2) # [b, cx?, t] -> [b, c, t, ?]
+
+    unnormalized_widths = h[..., :self.num_bins] / math.sqrt(self.filter_channels)
+    unnormalized_heights = h[..., self.num_bins:2*self.num_bins] / math.sqrt(self.filter_channels)
+    unnormalized_derivatives = h[..., 2 * self.num_bins:]
+
+    x1, logabsdet = piecewise_rational_quadratic_transform(x1,
+        unnormalized_widths,
+        unnormalized_heights,
+        unnormalized_derivatives,
+        inverse=reverse,
+        tails='linear',
+        tail_bound=self.tail_bound
+    )
+
+    x = torch.cat([x0, x1], 1) * x_mask
+    logdet = torch.sum(logabsdet * x_mask, [1,2])
+    if not reverse:
+        return x, logdet
+    else:
+        return x

monotonic_align/__init__.py

@@ -0,0 +1,19 @@
+import numpy as np
+import torch
+from .monotonic_align.core import maximum_path_c
+
+
+def maximum_path(neg_cent, mask):
+  """ Cython optimized version.
+  neg_cent: [b, t_t, t_s]
+  mask: [b, t_t, t_s]
+  """
+  device = neg_cent.device
+  dtype = neg_cent.dtype
+  neg_cent = neg_cent.data.cpu().numpy().astype(np.float32)
+  path = np.zeros(neg_cent.shape, dtype=np.int32)
+
+  t_t_max = mask.sum(1)[:, 0].data.cpu().numpy().astype(np.int32)
+  t_s_max = mask.sum(2)[:, 0].data.cpu().numpy().astype(np.int32)
+  maximum_path_c(path, neg_cent, t_t_max, t_s_max)
+  return torch.from_numpy(path).to(device=device, dtype=dtype)

monotonic_align/core.pyx

@@ -0,0 +1,42 @@
+cimport cython
+from cython.parallel import prange
+
+
+@cython.boundscheck(False)
+@cython.wraparound(False)
+cdef void maximum_path_each(int[:,::1] path, float[:,::1] value, int t_y, int t_x, float max_neg_val=-1e9) nogil:
+  cdef int x
+  cdef int y
+  cdef float v_prev
+  cdef float v_cur
+  cdef float tmp
+  cdef int index = t_x - 1
+
+  for y in range(t_y):
+    for x in range(max(0, t_x + y - t_y), min(t_x, y + 1)):
+      if x == y:
+        v_cur = max_neg_val
+      else:
+        v_cur = value[y-1, x]
+      if x == 0:
+        if y == 0:
+          v_prev = 0.
+        else:
+          v_prev = max_neg_val
+      else:
+        v_prev = value[y-1, x-1]
+      value[y, x] += max(v_prev, v_cur)
+
+  for y in range(t_y - 1, -1, -1):
+    path[y, index] = 1
+    if index != 0 and (index == y or value[y-1, index] < value[y-1, index-1]):
+      index = index - 1
+
+
+@cython.boundscheck(False)
+@cython.wraparound(False)
+cpdef void maximum_path_c(int[:,:,::1] paths, float[:,:,::1] values, int[::1] t_ys, int[::1] t_xs) nogil:
+  cdef int b = paths.shape[0]
+  cdef int i
+  for i in prange(b, nogil=True):
+    maximum_path_each(paths[i], values[i], t_ys[i], t_xs[i])

monotonic_align/setup.py

@@ -0,0 +1,9 @@
+from distutils.core import setup
+from Cython.Build import cythonize
+import numpy
+
+setup(
+  name = 'monotonic_align',
+  ext_modules = cythonize("core.pyx"),
+  include_dirs=[numpy.get_include()]
+)

preprocess.py

@@ -0,0 +1,25 @@
+import argparse
+import text
+from utils import load_filepaths_and_text
+
+if __name__ == '__main__':
+  parser = argparse.ArgumentParser()
+  parser.add_argument("--out_extension", default="cleaned")
+  parser.add_argument("--text_index", default=1, type=int)
+  parser.add_argument("--filelists", nargs="+", default=["filelists/ljs_audio_text_val_filelist.txt", "filelists/ljs_audio_text_test_filelist.txt"])
+  parser.add_argument("--text_cleaners", nargs="+", default=["english_cleaners2"])
+
+  args = parser.parse_args()
+    
+
+  for filelist in args.filelists:
+    print("START:", filelist)
+    filepaths_and_text = load_filepaths_and_text(filelist)
+    for i in range(len(filepaths_and_text)):
+      original_text = filepaths_and_text[i][args.text_index]
+      cleaned_text = text._clean_text(original_text, args.text_cleaners)
+      filepaths_and_text[i][args.text_index] = cleaned_text
+
+    new_filelist = filelist + "." + args.out_extension
+    with open(new_filelist, "w", encoding="utf-8") as f:
+      f.writelines(["|".join(x) + "\n" for x in filepaths_and_text])

requirements.txt

@@ -0,0 +1,11 @@
+Cython==0.29.21
+librosa==0.8.0
+matplotlib==3.3.1
+numpy==1.18.5
+phonemizer==2.2.1
+scipy==1.5.2
+tensorboard==2.3.0
+torch==1.6.0
+torchvision==0.7.0
+Unidecode==1.1.1
+torchtext

text/LICENSE

@@ -0,0 +1,19 @@
+Copyright (c) 2017 Keith Ito
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+THE SOFTWARE.

text/__init__.py

@@ -0,0 +1,54 @@
+""" from https://github.com/keithito/tacotron """
+from text import cleaners
+from text.symbols import symbols
+
+
+# Mappings from symbol to numeric ID and vice versa:
+_symbol_to_id = {s: i for i, s in enumerate(symbols)}
+_id_to_symbol = {i: s for i, s in enumerate(symbols)}
+
+
+def text_to_sequence(text, cleaner_names):
+  '''Converts a string of text to a sequence of IDs corresponding to the symbols in the text.
+    Args:
+      text: string to convert to a sequence
+      cleaner_names: names of the cleaner functions to run the text through
+    Returns:
+      List of integers corresponding to the symbols in the text
+  '''
+  sequence = []
+
+  clean_text = _clean_text(text, cleaner_names)
+  for symbol in clean_text:
+    symbol_id = _symbol_to_id[symbol]
+    sequence += [symbol_id]
+  return sequence
+
+
+def cleaned_text_to_sequence(cleaned_text):
+  '''Converts a string of text to a sequence of IDs corresponding to the symbols in the text.
+    Args:
+      text: string to convert to a sequence
+    Returns:
+      List of integers corresponding to the symbols in the text
+  '''
+  sequence = [_symbol_to_id[symbol] for symbol in cleaned_text]
+  return sequence
+
+
+def sequence_to_text(sequence):
+  '''Converts a sequence of IDs back to a string'''
+  result = ''
+  for symbol_id in sequence:
+    s = _id_to_symbol[symbol_id]
+    result += s
+  return result
+
+
+def _clean_text(text, cleaner_names):
+  for name in cleaner_names:
+    cleaner = getattr(cleaners, name)
+    if not cleaner:
+      raise Exception('Unknown cleaner: %s' % name)
+    text = cleaner(text)
+  return text

text/cleaners.py

@@ -0,0 +1,203 @@
+""" from https://github.com/keithito/tacotron """
+
+'''
+Cleaners are transformations that run over the input text at both training and eval time.
+
+Cleaners can be selected by passing a comma-delimited list of cleaner names as the "cleaners"
+hyperparameter. Some cleaners are English-specific. You'll typically want to use:
+  1. "english_cleaners" for English text
+  2. "transliteration_cleaners" for non-English text that can be transliterated to ASCII using
+     the Unidecode library (https://pypi.python.org/pypi/Unidecode)
+  3. "basic_cleaners" if you do not want to transliterate (in this case, you should also update
+     the symbols in symbols.py to match your data).
+'''
+
+import re
+from unidecode import unidecode
+import pyopenjtalk
+from janome.tokenizer import Tokenizer
+
+
+# Regular expression matching whitespace:
+_whitespace_re = re.compile(r'\s+')
+
+# List of (regular expression, replacement) pairs for abbreviations:
+_abbreviations = [(re.compile('\\b%s\\.' % x[0], re.IGNORECASE), x[1]) for x in [
+  ('mrs', 'misess'),
+  ('mr', 'mister'),
+  ('dr', 'doctor'),
+  ('st', 'saint'),
+  ('co', 'company'),
+  ('jr', 'junior'),
+  ('maj', 'major'),
+  ('gen', 'general'),
+  ('drs', 'doctors'),
+  ('rev', 'reverend'),
+  ('lt', 'lieutenant'),
+  ('hon', 'honorable'),
+  ('sgt', 'sergeant'),
+  ('capt', 'captain'),
+  ('esq', 'esquire'),
+  ('ltd', 'limited'),
+  ('col', 'colonel'),
+  ('ft', 'fort'),
+]]
+
+# Regular expression matching Japanese without punctuation marks:
+_japanese_characters = re.compile(r'[A-Za-z\d\u3005\u3040-\u30ff\u4e00-\u9fff\uff11-\uff19\uff21-\uff3a\uff41-\uff5a\uff66-\uff9d]')
+
+# Regular expression matching non-Japanese characters or punctuation marks:
+_japanese_marks = re.compile(r'[^A-Za-z\d\u3005\u3040-\u30ff\u4e00-\u9fff\uff11-\uff19\uff21-\uff3a\uff41-\uff5a\uff66-\uff9d]')
+
+
+# Tokenizer for Japanese
+tokenizer = Tokenizer()
+
+
+def expand_abbreviations(text):
+  for regex, replacement in _abbreviations:
+    text = re.sub(regex, replacement, text)
+  return text
+
+
+
+
+def lowercase(text):
+  return text.lower()
+
+
+def collapse_whitespace(text):
+  return re.sub(_whitespace_re, ' ', text)
+
+
+def convert_to_ascii(text):
+  return unidecode(text)
+
+
+def basic_cleaners(text):
+  '''Basic pipeline that lowercases and collapses whitespace without transliteration.'''
+  text = lowercase(text)
+  text = collapse_whitespace(text)
+  return text
+
+
+def transliteration_cleaners(text):
+  '''Pipeline for non-English text that transliterates to ASCII.'''
+  text = convert_to_ascii(text)
+  text = lowercase(text)
+  text = collapse_whitespace(text)
+  return text
+
+
+
+def japanese_cleaners(text):
+  '''Pipeline for Japanese text.'''
+  sentences = re.split(_japanese_marks, text)
+  marks = re.findall(_japanese_marks, text)
+  text = ''
+  for i, mark in enumerate(marks):
+    if re.match(_japanese_characters, sentences[i]):
+      text += pyopenjtalk.g2p(sentences[i], kana=False).replace('pau','').replace(' ','')
+    text += unidecode(mark).replace(' ','')
+  if re.match(_japanese_characters, sentences[-1]):
+      text += pyopenjtalk.g2p(sentences[-1], kana=False).replace('pau','').replace(' ','')
+  if re.match('[A-Za-z]',text[-1]):
+    text += '.'
+  return text
+
+
+def japanese_tokenization_cleaners(text):
+  '''Pipeline for tokenizing Japanese text.'''
+  words = []
+  for token in tokenizer.tokenize(text):
+    if token.phonetic!='*':
+      words.append(token.phonetic)
+    else:
+      words.append(token.surface)
+  text = ''
+  for word in words:
+    if re.match(_japanese_characters, word):
+      if word[0] == '\u30fc':
+        continue
+      if len(text)>0:
+        text += ' '
+      text += pyopenjtalk.g2p(word, kana=False).replace(' ','')
+    else:
+      text += unidecode(word).replace(' ','')
+  if re.match('[A-Za-z]',text[-1]):
+    text += '.'
+  return text
+
+
+def japanese_accent_cleaners(text):
+  '''Pipeline for notating accent in Japanese text.'''
+  '''Reference https://r9y9.github.io/ttslearn/latest/notebooks/ch10_Recipe-Tacotron.html'''
+  sentences = re.split(_japanese_marks, text)
+  marks = re.findall(_japanese_marks, text)
+  text = ''
+  for i, sentence in enumerate(sentences):
+    if re.match(_japanese_characters, sentence):
+      text += ':'
+      labels = pyopenjtalk.extract_fullcontext(sentence)
+      for n, label in enumerate(labels):
+        phoneme = re.search(r'\-([^\+]*)\+', label).group(1)
+        if phoneme not in ['sil','pau']:
+          text += phoneme
+        else:
+          continue
+        n_moras = int(re.search(r'/F:(\d+)_', label).group(1))
+        a1 = int(re.search(r"/A:(\-?[0-9]+)\+", label).group(1))
+        a2 = int(re.search(r"\+(\d+)\+", label).group(1))
+        a3 = int(re.search(r"\+(\d+)/", label).group(1))
+        if re.search(r'\-([^\+]*)\+', labels[n + 1]).group(1) in ['sil','pau']:
+          a2_next=-1
+        else:
+          a2_next = int(re.search(r"\+(\d+)\+", labels[n + 1]).group(1))
+        # Accent phrase boundary
+        if a3 == 1 and a2_next == 1:
+          text += ' '
+        # Falling
+        elif a1 == 0 and a2_next == a2 + 1 and a2 != n_moras:
+          text += ')'
+        # Rising
+        elif a2 == 1 and a2_next == 2:
+          text += '('
+    if i<len(marks):
+      text += unidecode(marks[i]).replace(' ','')
+  if re.match('[A-Za-z]',text[-1]):
+    text += '.'
+  return text
+
+
+def japanese_phrase_cleaners(text):
+  '''Pipeline for dividing Japanese text into phrases.'''
+  sentences = re.split(_japanese_marks, text)
+  marks = re.findall(_japanese_marks, text)
+  text = ''
+  for i, sentence in enumerate(sentences):
+    if re.match(_japanese_characters, sentence):
+      if text != '':
+        text += ' '
+      labels = pyopenjtalk.extract_fullcontext(sentence)
+      for n, label in enumerate(labels):
+        phoneme = re.search(r'\-([^\+]*)\+', label).group(1)
+        if phoneme not in ['sil','pau']:
+          text += phoneme
+        else:
+          continue
+        a3 = int(re.search(r"\+(\d+)/", label).group(1))
+        if re.search(r'\-([^\+]*)\+', labels[n + 1]).group(1) in ['sil','pau']:
+          a2_next=-1
+        else:
+          a2_next = int(re.search(r"\+(\d+)\+", labels[n + 1]).group(1))
+        # Accent phrase boundary
+        if a3 == 1 and a2_next == 1:
+          text += ' '
+    if i<len(marks):
+      text += unidecode(marks[i]).replace(' ','')
+  if re.match('[A-Za-z]',text[-1]):
+    text += '.'
+  return text
+
+
+

text/symbols.py

@@ -0,0 +1,2 @@
+symbols = list(' !"&*,-.?ABCINU[]abcdefghijklmnoprstuwyz{}~():=x')
+SPACE_ID = symbols.index(" ")

train.py

@@ -0,0 +1,304 @@
+import os
+import json
+import argparse
+import itertools
+import math
+import torch
+from torch import nn, optim
+from torch.nn import functional as F
+from torch.utils.data import DataLoader
+from torch.utils.tensorboard import SummaryWriter
+import torch.multiprocessing as mp
+import torch.distributed as dist
+from torch.nn.parallel import DistributedDataParallel as DDP
+from torch.cuda.amp import autocast, GradScaler
+
+import commons
+import utils
+from data_utils import (
+  TextAudioLoader,
+  TextAudioCollate,
+  DistributedBucketSampler
+)
+from models import (
+  SynthesizerTrn,
+  MultiPeriodDiscriminator,
+)
+from losses import (
+  generator_loss,
+  discriminator_loss,
+  feature_loss,
+  kl_loss
+)
+from mel_processing import mel_spectrogram_torch, spec_to_mel_torch
+from text.symbols import symbols
+
+
+torch.backends.cudnn.benchmark = True
+global_step = 0
+
+
+def main():
+  """Assume Single Node Multi GPUs Training Only"""
+  assert torch.cuda.is_available(), "CPU training is not allowed."
+
+  n_gpus = torch.cuda.device_count()
+  os.environ['MASTER_ADDR'] = 'localhost'
+  os.environ['MASTER_PORT'] = '8181'
+
+  hps = utils.get_hparams()
+  mp.spawn(run, nprocs=n_gpus, args=(n_gpus, hps,))
+
+
+def run(rank, n_gpus, hps):
+  global global_step
+  if rank == 0:
+    logger = utils.get_logger(hps.model_dir)
+    logger.info(hps)
+    utils.check_git_hash(hps.model_dir)
+    writer = SummaryWriter(log_dir=hps.model_dir)
+    writer_eval = SummaryWriter(log_dir=os.path.join(hps.model_dir, "eval"))
+
+  dist.init_process_group(backend='nccl', init_method='env://', world_size=n_gpus, rank=rank)
+  torch.manual_seed(hps.train.seed)
+  torch.cuda.set_device(rank)
+
+  train_dataset = TextAudioLoader(hps.data.training_files, hps.data)
+  train_sampler = DistributedBucketSampler(
+      train_dataset,
+      hps.train.batch_size,
+      [32,300,400,500,600,700,800,900,1000],
+      num_replicas=n_gpus,
+      rank=rank,
+      shuffle=True)
+  collate_fn = TextAudioCollate()
+  train_loader = DataLoader(train_dataset, num_workers=8, shuffle=False, pin_memory=True,
+      collate_fn=collate_fn, batch_sampler=train_sampler)
+  if rank == 0:
+    eval_dataset = TextAudioLoader(hps.data.validation_files, hps.data)
+    eval_loader = DataLoader(eval_dataset, num_workers=8, shuffle=False,
+        batch_size=hps.train.batch_size, pin_memory=True,
+        drop_last=False, collate_fn=collate_fn)
+
+  net_g = SynthesizerTrn(
+      len(symbols),
+      hps.data.filter_length // 2 + 1,
+      hps.train.segment_size // hps.data.hop_length,
+      **hps.model).cuda(rank)
+  net_d = MultiPeriodDiscriminator(hps.model.use_spectral_norm).cuda(rank)
+  optim_g = torch.optim.AdamW(
+      net_g.parameters(), 
+      hps.train.learning_rate, 
+      betas=hps.train.betas, 
+      eps=hps.train.eps)
+  optim_d = torch.optim.AdamW(
+      net_d.parameters(),
+      hps.train.learning_rate, 
+      betas=hps.train.betas, 
+      eps=hps.train.eps)
+  net_g = DDP(net_g, device_ids=[rank])
+  net_d = DDP(net_d, device_ids=[rank])
+
+  try:
+    _, _, _, epoch_str = utils.load_checkpoint(utils.latest_checkpoint_path(hps.model_dir, "G_*.pth"), net_g, optim_g)
+    _, _, _, epoch_str = utils.load_checkpoint(utils.latest_checkpoint_path(hps.model_dir, "D_*.pth"), net_d, optim_d)
+    global_step = (epoch_str - 1) * len(train_loader)
+  except:
+    epoch_str = 1
+    global_step = 0
+
+  scheduler_g = torch.optim.lr_scheduler.ExponentialLR(optim_g, gamma=hps.train.lr_decay, last_epoch=epoch_str-2)
+  scheduler_d = torch.optim.lr_scheduler.ExponentialLR(optim_d, gamma=hps.train.lr_decay, last_epoch=epoch_str-2)
+
+  scaler = GradScaler(enabled=hps.train.fp16_run)
+
+  for epoch in range(epoch_str, hps.train.epochs + 1):
+    if rank==0:
+      train_and_evaluate(rank, epoch, hps, [net_g, net_d], [optim_g, optim_d], [scheduler_g, scheduler_d], scaler, [train_loader, eval_loader], logger, [writer, writer_eval])
+    else:
+      train_and_evaluate(rank, epoch, hps, [net_g, net_d], [optim_g, optim_d], [scheduler_g, scheduler_d], scaler, [train_loader, None], None, None)
+    scheduler_g.step()
+    scheduler_d.step()
+
+
+def train_and_evaluate(rank, epoch, hps, nets, optims, schedulers, scaler, loaders, logger, writers):
+  net_g, net_d = nets
+  optim_g, optim_d = optims
+  scheduler_g, scheduler_d = schedulers
+  train_loader, eval_loader = loaders
+  if writers is not None:
+    writer, writer_eval = writers
+
+  train_loader.batch_sampler.set_epoch(epoch)
+  global global_step
+
+  net_g.train()
+  net_d.train()
+  for batch_idx, (x, x_lengths, spec, spec_lengths, y, y_lengths) in enumerate(train_loader):
+    step_save = False
+    x, x_lengths = x.cuda(rank, non_blocking=True), x_lengths.cuda(rank, non_blocking=True)
+    spec, spec_lengths = spec.cuda(rank, non_blocking=True), spec_lengths.cuda(rank, non_blocking=True)
+    y, y_lengths = y.cuda(rank, non_blocking=True), y_lengths.cuda(rank, non_blocking=True)
+    
+    with autocast(enabled=hps.train.fp16_run):
+      y_hat, l_length, attn, ids_slice, x_mask, z_mask,\
+      (z, z_p, m_p, logs_p, m_q, logs_q) = net_g(x, x_lengths, spec, spec_lengths)
+
+      mel = spec_to_mel_torch(
+          spec, 
+          hps.data.filter_length, 
+          hps.data.n_mel_channels, 
+          hps.data.sampling_rate,
+          hps.data.mel_fmin, 
+          hps.data.mel_fmax)
+      y_mel = commons.slice_segments(mel, ids_slice, hps.train.segment_size // hps.data.hop_length)
+      y_hat_mel = mel_spectrogram_torch(
+          y_hat.squeeze(1), 
+          hps.data.filter_length, 
+          hps.data.n_mel_channels, 
+          hps.data.sampling_rate, 
+          hps.data.hop_length, 
+          hps.data.win_length, 
+          hps.data.mel_fmin, 
+          hps.data.mel_fmax
+      )
+
+      y = commons.slice_segments(y, ids_slice * hps.data.hop_length, hps.train.segment_size) # slice 
+
+      # Discriminator
+      y_d_hat_r, y_d_hat_g, _, _ = net_d(y, y_hat.detach())
+      with autocast(enabled=False):
+        loss_disc, losses_disc_r, losses_disc_g = discriminator_loss(y_d_hat_r, y_d_hat_g)
+        loss_disc_all = loss_disc
+    optim_d.zero_grad()
+    scaler.scale(loss_disc_all).backward()
+    scaler.unscale_(optim_d)
+    grad_norm_d = commons.clip_grad_value_(net_d.parameters(), None)
+    scaler.step(optim_d)
+
+    with autocast(enabled=hps.train.fp16_run):
+      # Generator
+      y_d_hat_r, y_d_hat_g, fmap_r, fmap_g = net_d(y, y_hat)
+      with autocast(enabled=False):
+        loss_dur = torch.sum(l_length.float())
+        loss_mel = F.l1_loss(y_mel, y_hat_mel) * hps.train.c_mel
+        loss_kl = kl_loss(z_p, logs_q, m_p, logs_p, z_mask) * hps.train.c_kl
+
+        loss_fm = feature_loss(fmap_r, fmap_g)
+        loss_gen, losses_gen = generator_loss(y_d_hat_g)
+        loss_gen_all = loss_gen + loss_fm + loss_mel + loss_dur + loss_kl
+    optim_g.zero_grad()
+    scaler.scale(loss_gen_all).backward()
+    scaler.unscale_(optim_g)
+    grad_norm_g = commons.clip_grad_value_(net_g.parameters(), None)
+    scaler.step(optim_g)
+    scaler.update()
+
+    if rank==0:
+      if global_step % hps.train.log_interval == 0:
+        lr = optim_g.param_groups[0]['lr']
+        losses = [loss_disc, loss_gen, loss_fm, loss_mel, loss_dur, loss_kl]
+        logger.info('Train Epoch: {} [{:.0f}%]'.format(
+          epoch,
+          100. * batch_idx / len(train_loader)))
+        logger.info([x.item() for x in losses] + [global_step, lr])
+        
+        scalar_dict = {"loss/g/total": loss_gen_all, "loss/d/total": loss_disc_all, "learning_rate": lr, "grad_norm_d": grad_norm_d, "grad_norm_g": grad_norm_g}
+        scalar_dict.update({"loss/g/fm": loss_fm, "loss/g/mel": loss_mel, "loss/g/dur": loss_dur, "loss/g/kl": loss_kl})
+
+        scalar_dict.update({"loss/g/{}".format(i): v for i, v in enumerate(losses_gen)})
+        scalar_dict.update({"loss/d_r/{}".format(i): v for i, v in enumerate(losses_disc_r)})
+        scalar_dict.update({"loss/d_g/{}".format(i): v for i, v in enumerate(losses_disc_g)})
+        image_dict = { 
+            "slice/mel_org": utils.plot_spectrogram_to_numpy(y_mel[0].data.cpu().numpy()),
+            "slice/mel_gen": utils.plot_spectrogram_to_numpy(y_hat_mel[0].data.cpu().numpy()), 
+            "all/mel": utils.plot_spectrogram_to_numpy(mel[0].data.cpu().numpy()),
+            "all/attn": utils.plot_alignment_to_numpy(attn[0,0].data.cpu().numpy())
+        }
+        utils.summarize(
+          writer=writer,
+          global_step=global_step, 
+          images=image_dict,
+          scalars=scalar_dict)
+
+      if global_step % hps.train.eval_interval == 0:
+        evaluate(hps, net_g, eval_loader, writer_eval)
+        utils.save_checkpoint(net_g, optim_g, hps.train.learning_rate, epoch, os.path.join(hps.model_dir, "G_{}.pth".format(global_step)))
+        utils.save_checkpoint(net_d, optim_d, hps.train.learning_rate, epoch, os.path.join(hps.model_dir, "D_{}.pth".format(global_step)))
+        step_save = True
+      if global_step % hps.train.colab_save_interval == 0:
+        if step_save == False:
+          utils.save_checkpoint(net_g, optim_g, hps.train.learning_rate, epoch, os.path.join(hps.model_dir, "G_{}.pth".format(global_step)))
+          utils.save_checkpoint(net_d, optim_d, hps.train.learning_rate, epoch, os.path.join(hps.model_dir, "D_{}.pth".format(global_step)))
+        try:
+          os.mkdir(f'/content/gdrive/MyDrive/model/{hps.model_dir.split("/")[-1]}/')
+          os.system(f'cp {hps.model_dir}/G_{global_step}.pth /content/gdrive/MyDrive/model/{hps.model_dir.split("/")[-1]}/')
+          os.system(f'cp {hps.model_dir}/D_{global_step}.pth /content/gdrive/MyDrive/model/{hps.model_dir.split("/")[-1]}/')
+        except:
+          os.system(f'cp {hps.model_dir}/G_{global_step}.pth /content/gdrive/MyDrive/model/{hps.model_dir.split("/")[-1]}/')
+          os.system(f'cp {hps.model_dir}/D_{global_step}.pth /content/gdrive/MyDrive/model/{hps.model_dir.split("/")[-1]}/')
+        step_save = True
+    global_step += 1
+  
+  if rank == 0:
+    logger.info('====> Epoch: {}'.format(epoch))
+
+ 
+def evaluate(hps, generator, eval_loader, writer_eval):
+    generator.eval()
+    with torch.no_grad():
+      for batch_idx, (x, x_lengths, spec, spec_lengths, y, y_lengths) in enumerate(eval_loader):
+        x, x_lengths = x.cuda(0), x_lengths.cuda(0)
+        spec, spec_lengths = spec.cuda(0), spec_lengths.cuda(0)
+        y, y_lengths = y.cuda(0), y_lengths.cuda(0)
+
+        # remove else
+        x = x[:1]
+        x_lengths = x_lengths[:1]
+        spec = spec[:1]
+        spec_lengths = spec_lengths[:1]
+        y = y[:1]
+        y_lengths = y_lengths[:1]
+        break
+      y_hat, attn, mask, *_ = generator.module.infer(x, x_lengths, max_len=1000)
+      y_hat_lengths = mask.sum([1,2]).long() * hps.data.hop_length
+
+      mel = spec_to_mel_torch(
+        spec, 
+        hps.data.filter_length, 
+        hps.data.n_mel_channels, 
+        hps.data.sampling_rate,
+        hps.data.mel_fmin, 
+        hps.data.mel_fmax)
+      y_hat_mel = mel_spectrogram_torch(
+        y_hat.squeeze(1).float(),
+        hps.data.filter_length,
+        hps.data.n_mel_channels,
+        hps.data.sampling_rate,
+        hps.data.hop_length,
+        hps.data.win_length,
+        hps.data.mel_fmin,
+        hps.data.mel_fmax
+      )
+    image_dict = {
+      "gen/mel": utils.plot_spectrogram_to_numpy(y_hat_mel[0].cpu().numpy())
+    }
+    audio_dict = {
+      "gen/audio": y_hat[0,:,:y_hat_lengths[0]]
+    }
+    if global_step == 0:
+      image_dict.update({"gt/mel": utils.plot_spectrogram_to_numpy(mel[0].cpu().numpy())})
+      audio_dict.update({"gt/audio": y[0,:,:y_lengths[0]]})
+
+    utils.summarize(
+      writer=writer_eval,
+      global_step=global_step, 
+      images=image_dict,
+      audios=audio_dict,
+      audio_sampling_rate=hps.data.sampling_rate
+    )
+    generator.train()
+
+                           
+if __name__ == "__main__":
+  main()

train_ms.py

@@ -0,0 +1,308 @@
+import os
+import json
+import argparse
+import itertools
+import math
+import torch
+from torch import nn, optim
+from torch.nn import functional as F
+from torch.utils.data import DataLoader
+from torch.utils.tensorboard import SummaryWriter
+import torch.multiprocessing as mp
+import torch.distributed as dist
+from torch.nn.parallel import DistributedDataParallel as DDP
+from torch.cuda.amp import autocast, GradScaler
+
+import commons
+import utils
+from data_utils import (
+  TextAudioSpeakerLoader,
+  TextAudioSpeakerCollate,
+  DistributedBucketSampler
+)
+from models import (
+  SynthesizerTrn,
+  MultiPeriodDiscriminator,
+)
+from losses import (
+  generator_loss,
+  discriminator_loss,
+  feature_loss,
+  kl_loss
+)
+from mel_processing import mel_spectrogram_torch, spec_to_mel_torch
+from text.symbols import symbols
+
+
+torch.backends.cudnn.benchmark = True
+global_step = 0
+
+
+def main():
+  """Assume Single Node Multi GPUs Training Only"""
+  assert torch.cuda.is_available(), "CPU training is not allowed."
+
+  n_gpus = torch.cuda.device_count()
+  os.environ['MASTER_ADDR'] = 'localhost'
+  os.environ['MASTER_PORT'] = '8181'
+
+  hps = utils.get_hparams()
+  mp.spawn(run, nprocs=n_gpus, args=(n_gpus, hps,))
+
+
+def run(rank, n_gpus, hps):
+  global global_step
+  if rank == 0:
+    logger = utils.get_logger(hps.model_dir)
+    logger.info(hps)
+    utils.check_git_hash(hps.model_dir)
+    writer = SummaryWriter(log_dir=hps.model_dir)
+    writer_eval = SummaryWriter(log_dir=os.path.join(hps.model_dir, "eval"))
+
+  dist.init_process_group(backend='nccl', init_method='env://', world_size=n_gpus, rank=rank)
+  torch.manual_seed(hps.train.seed)
+  torch.cuda.set_device(rank)
+
+  train_dataset = TextAudioSpeakerLoader(hps.data.training_files, hps.data)
+  train_sampler = DistributedBucketSampler(
+      train_dataset,
+      hps.train.batch_size,
+      [32,300,400,500,600,700,800,900,1000],
+      num_replicas=n_gpus,
+      rank=rank,
+      shuffle=True)
+  collate_fn = TextAudioSpeakerCollate()
+  train_loader = DataLoader(train_dataset, num_workers=8, shuffle=False, pin_memory=True,
+      collate_fn=collate_fn, batch_sampler=train_sampler)
+  if rank == 0:
+    eval_dataset = TextAudioSpeakerLoader(hps.data.validation_files, hps.data)
+    eval_loader = DataLoader(eval_dataset, num_workers=8, shuffle=False,
+        batch_size=hps.train.batch_size, pin_memory=True,
+        drop_last=False, collate_fn=collate_fn)
+
+  net_g = SynthesizerTrn(
+      len(symbols),
+      hps.data.filter_length // 2 + 1,
+      hps.train.segment_size // hps.data.hop_length,
+      n_speakers=hps.data.n_speakers,
+      **hps.model).cuda(rank)
+  net_d = MultiPeriodDiscriminator(hps.model.use_spectral_norm).cuda(rank)
+  optim_g = torch.optim.AdamW(
+      net_g.parameters(), 
+      hps.train.learning_rate, 
+      betas=hps.train.betas, 
+      eps=hps.train.eps)
+  optim_d = torch.optim.AdamW(
+      net_d.parameters(),
+      hps.train.learning_rate, 
+      betas=hps.train.betas, 
+      eps=hps.train.eps)
+  net_g = DDP(net_g, device_ids=[rank])
+  net_d = DDP(net_d, device_ids=[rank])
+
+  try:
+    _, _, _, epoch_str = utils.load_checkpoint(utils.latest_checkpoint_path(hps.model_dir, "G_*.pth"), net_g, optim_g)
+    _, _, _, epoch_str = utils.load_checkpoint(utils.latest_checkpoint_path(hps.model_dir, "D_*.pth"), net_d, optim_d)
+    global_step = (epoch_str - 1) * len(train_loader)
+  except:
+    epoch_str = 1
+    global_step = 0
+
+  scheduler_g = torch.optim.lr_scheduler.ExponentialLR(optim_g, gamma=hps.train.lr_decay, last_epoch=epoch_str-2)
+  scheduler_d = torch.optim.lr_scheduler.ExponentialLR(optim_d, gamma=hps.train.lr_decay, last_epoch=epoch_str-2)
+
+  scaler = GradScaler(enabled=hps.train.fp16_run)
+
+  for epoch in range(epoch_str, hps.train.epochs + 1):
+    if rank==0:
+      train_and_evaluate(rank, epoch, hps, [net_g, net_d], [optim_g, optim_d], [scheduler_g, scheduler_d], scaler, [train_loader, eval_loader], logger, [writer, writer_eval])
+    else:
+      train_and_evaluate(rank, epoch, hps, [net_g, net_d], [optim_g, optim_d], [scheduler_g, scheduler_d], scaler, [train_loader, None], None, None)
+    scheduler_g.step()
+    scheduler_d.step()
+
+
+def train_and_evaluate(rank, epoch, hps, nets, optims, schedulers, scaler, loaders, logger, writers):
+  net_g, net_d = nets
+  optim_g, optim_d = optims
+  scheduler_g, scheduler_d = schedulers
+  train_loader, eval_loader = loaders
+  if writers is not None:
+    writer, writer_eval = writers
+
+  train_loader.batch_sampler.set_epoch(epoch)
+  global global_step
+
+  net_g.train()
+  net_d.train()
+  for batch_idx, (x, x_lengths, spec, spec_lengths, y, y_lengths, speakers) in enumerate(train_loader):
+    step_save = False
+    x, x_lengths = x.cuda(rank, non_blocking=True), x_lengths.cuda(rank, non_blocking=True)
+    spec, spec_lengths = spec.cuda(rank, non_blocking=True), spec_lengths.cuda(rank, non_blocking=True)
+    y, y_lengths = y.cuda(rank, non_blocking=True), y_lengths.cuda(rank, non_blocking=True)
+    speakers = speakers.cuda(rank, non_blocking=True)
+
+    with autocast(enabled=hps.train.fp16_run):
+      y_hat, l_length, attn, ids_slice, x_mask, z_mask,\
+      (z, z_p, m_p, logs_p, m_q, logs_q) = net_g(x, x_lengths, spec, spec_lengths, speakers)
+
+      mel = spec_to_mel_torch(
+          spec, 
+          hps.data.filter_length, 
+          hps.data.n_mel_channels, 
+          hps.data.sampling_rate,
+          hps.data.mel_fmin, 
+          hps.data.mel_fmax)
+      y_mel = commons.slice_segments(mel, ids_slice, hps.train.segment_size // hps.data.hop_length)
+      y_hat_mel = mel_spectrogram_torch(
+          y_hat.squeeze(1), 
+          hps.data.filter_length, 
+          hps.data.n_mel_channels, 
+          hps.data.sampling_rate, 
+          hps.data.hop_length, 
+          hps.data.win_length, 
+          hps.data.mel_fmin, 
+          hps.data.mel_fmax
+      )
+
+      y = commons.slice_segments(y, ids_slice * hps.data.hop_length, hps.train.segment_size) # slice 
+
+      # Discriminator
+      y_d_hat_r, y_d_hat_g, _, _ = net_d(y, y_hat.detach())
+      with autocast(enabled=False):
+        loss_disc, losses_disc_r, losses_disc_g = discriminator_loss(y_d_hat_r, y_d_hat_g)
+        loss_disc_all = loss_disc
+    optim_d.zero_grad()
+    scaler.scale(loss_disc_all).backward()
+    scaler.unscale_(optim_d)
+    grad_norm_d = commons.clip_grad_value_(net_d.parameters(), None)
+    scaler.step(optim_d)
+
+    with autocast(enabled=hps.train.fp16_run):
+      # Generator
+      y_d_hat_r, y_d_hat_g, fmap_r, fmap_g = net_d(y, y_hat)
+      with autocast(enabled=False):
+        loss_dur = torch.sum(l_length.float())
+        loss_mel = F.l1_loss(y_mel, y_hat_mel) * hps.train.c_mel
+        loss_kl = kl_loss(z_p, logs_q, m_p, logs_p, z_mask) * hps.train.c_kl
+
+        loss_fm = feature_loss(fmap_r, fmap_g)
+        loss_gen, losses_gen = generator_loss(y_d_hat_g)
+        loss_gen_all = loss_gen + loss_fm + loss_mel + loss_dur + loss_kl
+    optim_g.zero_grad()
+    scaler.scale(loss_gen_all).backward()
+    scaler.unscale_(optim_g)
+    grad_norm_g = commons.clip_grad_value_(net_g.parameters(), None)
+    scaler.step(optim_g)
+    scaler.update()
+
+    if rank==0:
+      if global_step % hps.train.log_interval == 0:
+        lr = optim_g.param_groups[0]['lr']
+        losses = [loss_disc, loss_gen, loss_fm, loss_mel, loss_dur, loss_kl]
+        logger.info('Train Epoch: {} [{:.0f}%]'.format(
+          epoch,
+          100. * batch_idx / len(train_loader)))
+        logger.info([x.item() for x in losses] + [global_step, lr])
+        
+        scalar_dict = {"loss/g/total": loss_gen_all, "loss/d/total": loss_disc_all, "learning_rate": lr, "grad_norm_d": grad_norm_d, "grad_norm_g": grad_norm_g}
+        scalar_dict.update({"loss/g/fm": loss_fm, "loss/g/mel": loss_mel, "loss/g/dur": loss_dur, "loss/g/kl": loss_kl})
+
+        scalar_dict.update({"loss/g/{}".format(i): v for i, v in enumerate(losses_gen)})
+        scalar_dict.update({"loss/d_r/{}".format(i): v for i, v in enumerate(losses_disc_r)})
+        scalar_dict.update({"loss/d_g/{}".format(i): v for i, v in enumerate(losses_disc_g)})
+        image_dict = { 
+            "slice/mel_org": utils.plot_spectrogram_to_numpy(y_mel[0].data.cpu().numpy()),
+            "slice/mel_gen": utils.plot_spectrogram_to_numpy(y_hat_mel[0].data.cpu().numpy()), 
+            "all/mel": utils.plot_spectrogram_to_numpy(mel[0].data.cpu().numpy()),
+            "all/attn": utils.plot_alignment_to_numpy(attn[0,0].data.cpu().numpy())
+        }
+        utils.summarize(
+          writer=writer,
+          global_step=global_step, 
+          images=image_dict,
+          scalars=scalar_dict)
+
+      if global_step % hps.train.eval_interval == 0:
+        evaluate(hps, net_g, eval_loader, writer_eval)
+        utils.save_checkpoint(net_g, optim_g, hps.train.learning_rate, epoch, os.path.join(hps.model_dir, "G_{}.pth".format(global_step)))
+        utils.save_checkpoint(net_d, optim_d, hps.train.learning_rate, epoch, os.path.join(hps.model_dir, "D_{}.pth".format(global_step)))
+        step_save = True
+      if global_step % hps.train.colab_save_interval == 0:
+        if step_save == False:
+          utils.save_checkpoint(net_g, optim_g, hps.train.learning_rate, epoch, os.path.join(hps.model_dir, "G_{}.pth".format(global_step)))
+          utils.save_checkpoint(net_d, optim_d, hps.train.learning_rate, epoch, os.path.join(hps.model_dir, "D_{}.pth".format(global_step)))
+        try:
+          os.mkdir(f'/content/gdrive/MyDrive/model/{hps.model_dir.split("/")[-1]}/')
+          os.system(f'cp {hps.model_dir}/G_{global_step}.pth /content/gdrive/MyDrive/model/{hps.model_dir.split("/")[-1]}/')
+          os.system(f'cp {hps.model_dir}/D_{global_step}.pth /content/gdrive/MyDrive/model/{hps.model_dir.split("/")[-1]}/')
+        except:
+          os.system(f'cp {hps.model_dir}/G_{global_step}.pth /content/gdrive/MyDrive/model/{hps.model_dir.split("/")[-1]}/')
+          os.system(f'cp {hps.model_dir}/D_{global_step}.pth /content/gdrive/MyDrive/model/{hps.model_dir.split("/")[-1]}/')
+        step_save = True
+    global_step += 1
+  
+  if rank == 0:
+    logger.info('====> Epoch: {}'.format(epoch))
+
+ 
+def evaluate(hps, generator, eval_loader, writer_eval):
+    generator.eval()
+    with torch.no_grad():
+      for batch_idx, (x, x_lengths, spec, spec_lengths, y, y_lengths, speakers) in enumerate(eval_loader):
+        x, x_lengths = x.cuda(0), x_lengths.cuda(0)
+        spec, spec_lengths = spec.cuda(0), spec_lengths.cuda(0)
+        y, y_lengths = y.cuda(0), y_lengths.cuda(0)
+        speakers = speakers.cuda(0)
+
+        # remove else
+        x = x[:1]
+        x_lengths = x_lengths[:1]
+        spec = spec[:1]
+        spec_lengths = spec_lengths[:1]
+        y = y[:1]
+        y_lengths = y_lengths[:1]
+        speakers = speakers[:1]
+        break
+      y_hat, attn, mask, *_ = generator.module.infer(x, x_lengths, speakers, max_len=1000)
+      y_hat_lengths = mask.sum([1,2]).long() * hps.data.hop_length
+
+      mel = spec_to_mel_torch(
+        spec, 
+        hps.data.filter_length, 
+        hps.data.n_mel_channels, 
+        hps.data.sampling_rate,
+        hps.data.mel_fmin, 
+        hps.data.mel_fmax)
+      y_hat_mel = mel_spectrogram_torch(
+        y_hat.squeeze(1).float(),
+        hps.data.filter_length,
+        hps.data.n_mel_channels,
+        hps.data.sampling_rate,
+        hps.data.hop_length,
+        hps.data.win_length,
+        hps.data.mel_fmin,
+        hps.data.mel_fmax
+      )
+    image_dict = {
+      "gen/mel": utils.plot_spectrogram_to_numpy(y_hat_mel[0].cpu().numpy())
+    }
+    audio_dict = {
+      "gen/audio": y_hat[0,:,:y_hat_lengths[0]]
+    }
+    if global_step == 0:
+      image_dict.update({"gt/mel": utils.plot_spectrogram_to_numpy(mel[0].cpu().numpy())})
+      audio_dict.update({"gt/audio": y[0,:,:y_lengths[0]]})
+
+    utils.summarize(
+      writer=writer_eval,
+      global_step=global_step, 
+      images=image_dict,
+      audios=audio_dict,
+      audio_sampling_rate=hps.data.sampling_rate
+    )
+    generator.train()
+
+                           
+if __name__ == "__main__":
+  main()

transforms.py

@@ -0,0 +1,193 @@
+import torch
+from torch.nn import functional as F
+
+import numpy as np
+
+
+DEFAULT_MIN_BIN_WIDTH = 1e-3
+DEFAULT_MIN_BIN_HEIGHT = 1e-3
+DEFAULT_MIN_DERIVATIVE = 1e-3
+
+
+def piecewise_rational_quadratic_transform(inputs, 
+                                           unnormalized_widths,
+                                           unnormalized_heights,
+                                           unnormalized_derivatives,
+                                           inverse=False,
+                                           tails=None, 
+                                           tail_bound=1.,
+                                           min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+                                           min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+                                           min_derivative=DEFAULT_MIN_DERIVATIVE):
+
+    if tails is None:
+        spline_fn = rational_quadratic_spline
+        spline_kwargs = {}
+    else:
+        spline_fn = unconstrained_rational_quadratic_spline
+        spline_kwargs = {
+            'tails': tails,
+            'tail_bound': tail_bound
+        }
+
+    outputs, logabsdet = spline_fn(
+            inputs=inputs,
+            unnormalized_widths=unnormalized_widths,
+            unnormalized_heights=unnormalized_heights,
+            unnormalized_derivatives=unnormalized_derivatives,
+            inverse=inverse,
+            min_bin_width=min_bin_width,
+            min_bin_height=min_bin_height,
+            min_derivative=min_derivative,
+            **spline_kwargs
+    )
+    return outputs, logabsdet
+
+
+def searchsorted(bin_locations, inputs, eps=1e-6):
+    bin_locations[..., -1] += eps
+    return torch.sum(
+        inputs[..., None] >= bin_locations,
+        dim=-1
+    ) - 1
+
+
+def unconstrained_rational_quadratic_spline(inputs,
+                                            unnormalized_widths,
+                                            unnormalized_heights,
+                                            unnormalized_derivatives,
+                                            inverse=False,
+                                            tails='linear',
+                                            tail_bound=1.,
+                                            min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+                                            min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+                                            min_derivative=DEFAULT_MIN_DERIVATIVE):
+    inside_interval_mask = (inputs >= -tail_bound) & (inputs <= tail_bound)
+    outside_interval_mask = ~inside_interval_mask
+
+    outputs = torch.zeros_like(inputs)
+    logabsdet = torch.zeros_like(inputs)
+
+    if tails == 'linear':
+        unnormalized_derivatives = F.pad(unnormalized_derivatives, pad=(1, 1))
+        constant = np.log(np.exp(1 - min_derivative) - 1)
+        unnormalized_derivatives[..., 0] = constant
+        unnormalized_derivatives[..., -1] = constant
+
+        outputs[outside_interval_mask] = inputs[outside_interval_mask]
+        logabsdet[outside_interval_mask] = 0
+    else:
+        raise RuntimeError('{} tails are not implemented.'.format(tails))
+
+    outputs[inside_interval_mask], logabsdet[inside_interval_mask] = rational_quadratic_spline(
+        inputs=inputs[inside_interval_mask],
+        unnormalized_widths=unnormalized_widths[inside_interval_mask, :],
+        unnormalized_heights=unnormalized_heights[inside_interval_mask, :],
+        unnormalized_derivatives=unnormalized_derivatives[inside_interval_mask, :],
+        inverse=inverse,
+        left=-tail_bound, right=tail_bound, bottom=-tail_bound, top=tail_bound,
+        min_bin_width=min_bin_width,
+        min_bin_height=min_bin_height,
+        min_derivative=min_derivative
+    )
+
+    return outputs, logabsdet
+
+def rational_quadratic_spline(inputs,
+                              unnormalized_widths,
+                              unnormalized_heights,
+                              unnormalized_derivatives,
+                              inverse=False,
+                              left=0., right=1., bottom=0., top=1.,
+                              min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+                              min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+                              min_derivative=DEFAULT_MIN_DERIVATIVE):
+    if torch.min(inputs) < left or torch.max(inputs) > right:
+        raise ValueError('Input to a transform is not within its domain')
+
+    num_bins = unnormalized_widths.shape[-1]
+
+    if min_bin_width * num_bins > 1.0:
+        raise ValueError('Minimal bin width too large for the number of bins')
+    if min_bin_height * num_bins > 1.0:
+        raise ValueError('Minimal bin height too large for the number of bins')
+
+    widths = F.softmax(unnormalized_widths, dim=-1)
+    widths = min_bin_width + (1 - min_bin_width * num_bins) * widths
+    cumwidths = torch.cumsum(widths, dim=-1)
+    cumwidths = F.pad(cumwidths, pad=(1, 0), mode='constant', value=0.0)
+    cumwidths = (right - left) * cumwidths + left
+    cumwidths[..., 0] = left
+    cumwidths[..., -1] = right
+    widths = cumwidths[..., 1:] - cumwidths[..., :-1]
+
+    derivatives = min_derivative + F.softplus(unnormalized_derivatives)
+
+    heights = F.softmax(unnormalized_heights, dim=-1)
+    heights = min_bin_height + (1 - min_bin_height * num_bins) * heights
+    cumheights = torch.cumsum(heights, dim=-1)
+    cumheights = F.pad(cumheights, pad=(1, 0), mode='constant', value=0.0)
+    cumheights = (top - bottom) * cumheights + bottom
+    cumheights[..., 0] = bottom
+    cumheights[..., -1] = top
+    heights = cumheights[..., 1:] - cumheights[..., :-1]
+
+    if inverse:
+        bin_idx = searchsorted(cumheights, inputs)[..., None]
+    else:
+        bin_idx = searchsorted(cumwidths, inputs)[..., None]
+
+    input_cumwidths = cumwidths.gather(-1, bin_idx)[..., 0]
+    input_bin_widths = widths.gather(-1, bin_idx)[..., 0]
+
+    input_cumheights = cumheights.gather(-1, bin_idx)[..., 0]
+    delta = heights / widths
+    input_delta = delta.gather(-1, bin_idx)[..., 0]
+
+    input_derivatives = derivatives.gather(-1, bin_idx)[..., 0]
+    input_derivatives_plus_one = derivatives[..., 1:].gather(-1, bin_idx)[..., 0]
+
+    input_heights = heights.gather(-1, bin_idx)[..., 0]
+
+    if inverse:
+        a = (((inputs - input_cumheights) * (input_derivatives
+                                             + input_derivatives_plus_one
+                                             - 2 * input_delta)
+              + input_heights * (input_delta - input_derivatives)))
+        b = (input_heights * input_derivatives
+             - (inputs - input_cumheights) * (input_derivatives
+                                              + input_derivatives_plus_one
+                                              - 2 * input_delta))
+        c = - input_delta * (inputs - input_cumheights)
+
+        discriminant = b.pow(2) - 4 * a * c
+        assert (discriminant >= 0).all()
+
+        root = (2 * c) / (-b - torch.sqrt(discriminant))
+        outputs = root * input_bin_widths + input_cumwidths
+
+        theta_one_minus_theta = root * (1 - root)
+        denominator = input_delta + ((input_derivatives + input_derivatives_plus_one - 2 * input_delta)
+                                     * theta_one_minus_theta)
+        derivative_numerator = input_delta.pow(2) * (input_derivatives_plus_one * root.pow(2)
+                                                     + 2 * input_delta * theta_one_minus_theta
+                                                     + input_derivatives * (1 - root).pow(2))
+        logabsdet = torch.log(derivative_numerator) - 2 * torch.log(denominator)
+
+        return outputs, -logabsdet
+    else:
+        theta = (inputs - input_cumwidths) / input_bin_widths
+        theta_one_minus_theta = theta * (1 - theta)
+
+        numerator = input_heights * (input_delta * theta.pow(2)
+                                     + input_derivatives * theta_one_minus_theta)
+        denominator = input_delta + ((input_derivatives + input_derivatives_plus_one - 2 * input_delta)
+                                     * theta_one_minus_theta)
+        outputs = input_cumheights + numerator / denominator
+
+        derivative_numerator = input_delta.pow(2) * (input_derivatives_plus_one * theta.pow(2)
+                                                     + 2 * input_delta * theta_one_minus_theta
+                                                     + input_derivatives * (1 - theta).pow(2))
+        logabsdet = torch.log(derivative_numerator) - 2 * torch.log(denominator)
+
+        return outputs, logabsdet

utils.py

@@ -0,0 +1,258 @@
+import os
+import glob
+import sys
+import argparse
+import logging
+import json
+import subprocess
+import numpy as np
+from scipy.io.wavfile import read
+import torch
+
+MATPLOTLIB_FLAG = False
+
+logging.basicConfig(stream=sys.stdout, level=logging.DEBUG)
+logger = logging
+
+
+def load_checkpoint(checkpoint_path, model, optimizer=None):
+  assert os.path.isfile(checkpoint_path)
+  checkpoint_dict = torch.load(checkpoint_path, map_location='cpu')
+  iteration = checkpoint_dict['iteration']
+  learning_rate = checkpoint_dict['learning_rate']
+  if optimizer is not None:
+    optimizer.load_state_dict(checkpoint_dict['optimizer'])
+  saved_state_dict = checkpoint_dict['model']
+  if hasattr(model, 'module'):
+    state_dict = model.module.state_dict()
+  else:
+    state_dict = model.state_dict()
+  new_state_dict= {}
+  for k, v in state_dict.items():
+    try:
+      new_state_dict[k] = saved_state_dict[k]
+    except:
+      logger.info("%s is not in the checkpoint" % k)
+      new_state_dict[k] = v
+  if hasattr(model, 'module'):
+    model.module.load_state_dict(new_state_dict)
+  else:
+    model.load_state_dict(new_state_dict)
+  logger.info("Loaded checkpoint '{}' (iteration {})" .format(
+    checkpoint_path, iteration))
+  return model, optimizer, learning_rate, iteration
+
+
+def save_checkpoint(model, optimizer, learning_rate, iteration, checkpoint_path):
+  logger.info("Saving model and optimizer state at iteration {} to {}".format(
+    iteration, checkpoint_path))
+  if hasattr(model, 'module'):
+    state_dict = model.module.state_dict()
+  else:
+    state_dict = model.state_dict()
+  torch.save({'model': state_dict,
+              'iteration': iteration,
+              'optimizer': optimizer.state_dict(),
+              'learning_rate': learning_rate}, checkpoint_path)
+
+
+def summarize(writer, global_step, scalars={}, histograms={}, images={}, audios={}, audio_sampling_rate=22050):
+  for k, v in scalars.items():
+    writer.add_scalar(k, v, global_step)
+  for k, v in histograms.items():
+    writer.add_histogram(k, v, global_step)
+  for k, v in images.items():
+    writer.add_image(k, v, global_step, dataformats='HWC')
+  for k, v in audios.items():
+    writer.add_audio(k, v, global_step, audio_sampling_rate)
+
+
+def latest_checkpoint_path(dir_path, regex="G_*.pth"):
+  f_list = glob.glob(os.path.join(dir_path, regex))
+  f_list.sort(key=lambda f: int("".join(filter(str.isdigit, f))))
+  x = f_list[-1]
+  print(x)
+  return x
+
+
+def plot_spectrogram_to_numpy(spectrogram):
+  global MATPLOTLIB_FLAG
+  if not MATPLOTLIB_FLAG:
+    import matplotlib
+    matplotlib.use("Agg")
+    MATPLOTLIB_FLAG = True
+    mpl_logger = logging.getLogger('matplotlib')
+    mpl_logger.setLevel(logging.WARNING)
+  import matplotlib.pylab as plt
+  import numpy as np
+  
+  fig, ax = plt.subplots(figsize=(10,2))
+  im = ax.imshow(spectrogram, aspect="auto", origin="lower",
+                  interpolation='none')
+  plt.colorbar(im, ax=ax)
+  plt.xlabel("Frames")
+  plt.ylabel("Channels")
+  plt.tight_layout()
+
+  fig.canvas.draw()
+  data = np.fromstring(fig.canvas.tostring_rgb(), dtype=np.uint8, sep='')
+  data = data.reshape(fig.canvas.get_width_height()[::-1] + (3,))
+  plt.close()
+  return data
+
+
+def plot_alignment_to_numpy(alignment, info=None):
+  global MATPLOTLIB_FLAG
+  if not MATPLOTLIB_FLAG:
+    import matplotlib
+    matplotlib.use("Agg")
+    MATPLOTLIB_FLAG = True
+    mpl_logger = logging.getLogger('matplotlib')
+    mpl_logger.setLevel(logging.WARNING)
+  import matplotlib.pylab as plt
+  import numpy as np
+
+  fig, ax = plt.subplots(figsize=(6, 4))
+  im = ax.imshow(alignment.transpose(), aspect='auto', origin='lower',
+                  interpolation='none')
+  fig.colorbar(im, ax=ax)
+  xlabel = 'Decoder timestep'
+  if info is not None:
+      xlabel += '\n\n' + info
+  plt.xlabel(xlabel)
+  plt.ylabel('Encoder timestep')
+  plt.tight_layout()
+
+  fig.canvas.draw()
+  data = np.fromstring(fig.canvas.tostring_rgb(), dtype=np.uint8, sep='')
+  data = data.reshape(fig.canvas.get_width_height()[::-1] + (3,))
+  plt.close()
+  return data
+
+
+def load_wav_to_torch(full_path):
+  sampling_rate, data = read(full_path)
+  return torch.FloatTensor(data.astype(np.float32)), sampling_rate
+
+
+def load_filepaths_and_text(filename, split="|"):
+  with open(filename, encoding='utf-8') as f:
+    filepaths_and_text = [line.strip().split(split) for line in f]
+  return filepaths_and_text
+
+
+def get_hparams(init=True):
+  parser = argparse.ArgumentParser()
+  parser.add_argument('-c', '--config', type=str, default="./configs/base.json",
+                      help='JSON file for configuration')
+  parser.add_argument('-m', '--model', type=str, required=True,
+                      help='Model name')
+  
+  args = parser.parse_args()
+  model_dir = os.path.join("./logs", args.model)
+
+  if not os.path.exists(model_dir):
+    os.makedirs(model_dir)
+
+  config_path = args.config
+  config_save_path = os.path.join(model_dir, "config.json")
+  if init:
+    with open(config_path, "r") as f:
+      data = f.read()
+    with open(config_save_path, "w") as f:
+      f.write(data)
+  else:
+    with open(config_save_path, "r") as f:
+      data = f.read()
+  config = json.loads(data)
+  
+  hparams = HParams(**config)
+  hparams.model_dir = model_dir
+  return hparams
+
+
+def get_hparams_from_dir(model_dir):
+  config_save_path = os.path.join(model_dir, "config.json")
+  with open(config_save_path, "r") as f:
+    data = f.read()
+  config = json.loads(data)
+
+  hparams =HParams(**config)
+  hparams.model_dir = model_dir
+  return hparams
+
+
+def get_hparams_from_file(config_path):
+  with open(config_path, "r") as f:
+    data = f.read()
+  config = json.loads(data)
+
+  hparams =HParams(**config)
+  return hparams
+
+
+def check_git_hash(model_dir):
+  source_dir = os.path.dirname(os.path.realpath(__file__))
+  if not os.path.exists(os.path.join(source_dir, ".git")):
+    logger.warn("{} is not a git repository, therefore hash value comparison will be ignored.".format(
+      source_dir
+    ))
+    return
+
+  cur_hash = subprocess.getoutput("git rev-parse HEAD")
+
+  path = os.path.join(model_dir, "githash")
+  if os.path.exists(path):
+    saved_hash = open(path).read()
+    if saved_hash != cur_hash:
+      logger.warn("git hash values are different. {}(saved) != {}(current)".format(
+        saved_hash[:8], cur_hash[:8]))
+  else:
+    open(path, "w").write(cur_hash)
+
+
+def get_logger(model_dir, filename="train.log"):
+  global logger
+  logger = logging.getLogger(os.path.basename(model_dir))
+  logger.setLevel(logging.DEBUG)
+  
+  formatter = logging.Formatter("%(asctime)s\t%(name)s\t%(levelname)s\t%(message)s")
+  if not os.path.exists(model_dir):
+    os.makedirs(model_dir)
+  h = logging.FileHandler(os.path.join(model_dir, filename))
+  h.setLevel(logging.DEBUG)
+  h.setFormatter(formatter)
+  logger.addHandler(h)
+  return logger
+
+
+class HParams():
+  def __init__(self, **kwargs):
+    for k, v in kwargs.items():
+      if type(v) == dict:
+        v = HParams(**v)
+      self[k] = v
+    
+  def keys(self):
+    return self.__dict__.keys()
+
+  def items(self):
+    return self.__dict__.items()
+
+  def values(self):
+    return self.__dict__.values()
+
+  def __len__(self):
+    return len(self.__dict__)
+
+  def __getitem__(self, key):
+    return getattr(self, key)
+
+  def __setitem__(self, key, value):
+    return setattr(self, key, value)
+
+  def __contains__(self, key):
+    return key in self.__dict__
+
+  def __repr__(self):
+    return self.__dict__.__repr__()