add files used by space
Browse files- MeanAudio.py +147 -0
- app.py +421 -0
- easyinfer.py +3 -0
- requirements.txt +27 -0
MeanAudio.py
ADDED
@@ -0,0 +1,147 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import warnings
|
2 |
+
warnings.filterwarnings("ignore", category=FutureWarning)
|
3 |
+
import logging
|
4 |
+
from pathlib import Path
|
5 |
+
import torch
|
6 |
+
import torchaudio
|
7 |
+
from meanaudio.eval_utils import (ModelConfig, all_model_cfg, generate_mf, generate_fm, setup_eval_logging)
|
8 |
+
from meanaudio.model.flow_matching import FlowMatching
|
9 |
+
from meanaudio.model.mean_flow import MeanFlow
|
10 |
+
from meanaudio.model.networks import MeanAudio, get_mean_audio
|
11 |
+
from meanaudio.model.utils.features_utils import FeaturesUtils
|
12 |
+
from huggingface_hub import snapshot_download
|
13 |
+
|
14 |
+
torch.backends.cuda.matmul.allow_tf32 = True
|
15 |
+
torch.backends.cudnn.allow_tf32 = True
|
16 |
+
log = logging.getLogger()
|
17 |
+
|
18 |
+
@torch.inference_mode()
|
19 |
+
def MeanAudioInference(
|
20 |
+
prompt='',
|
21 |
+
negative_prompt='',
|
22 |
+
model_path='',
|
23 |
+
encoder_name='t5_clap',
|
24 |
+
variant='meanaudio_mf',
|
25 |
+
duration=10,
|
26 |
+
cfg_strength=4.5,
|
27 |
+
num_steps=1,
|
28 |
+
output='./output',
|
29 |
+
seed=42,
|
30 |
+
full_precision=False,
|
31 |
+
use_rope=True,
|
32 |
+
text_c_dim=512,
|
33 |
+
use_meanflow=False
|
34 |
+
):
|
35 |
+
'''
|
36 |
+
prompt (str):
|
37 |
+
The text description guiding the audio generation (e.g., "a dog is barking").
|
38 |
+
negative_prompt (str):
|
39 |
+
A text description for sounds that should be avoided in the generated audio.
|
40 |
+
model_path (str):
|
41 |
+
Path to the model weights file. If empty, it defaults to ./weights/{variant}.pth.
|
42 |
+
encoder_name (str):
|
43 |
+
Specifies the text encoder to use (default: 't5_clap').
|
44 |
+
variant (str):
|
45 |
+
Specifies the model variant to load (default: 'meanaudio_mf'). Must be a key in all_model_cfg.
|
46 |
+
duration (int):
|
47 |
+
The desired duration of the generated audio in seconds (default: 10).
|
48 |
+
cfg_strength (float):
|
49 |
+
Classifier-Free Guidance strength. Ignored if use_meanflow is True or variant is 'meanaudio_mf' (default: 4.5).
|
50 |
+
num_steps (int):
|
51 |
+
Number of steps for the generation process (default: 1).
|
52 |
+
output (str):
|
53 |
+
Directory path where the generated audio file will be saved (default: './output').
|
54 |
+
seed (int):
|
55 |
+
Random seed for generation reproducibility (default: 42).
|
56 |
+
full_precision (bool):
|
57 |
+
If True, uses torch.float32 precision; otherwise, uses torch.bfloat16 (default: False).
|
58 |
+
use_rope (bool):
|
59 |
+
Whether to use Rotary Position Embedding in the model (default: True).
|
60 |
+
text_c_dim (int):
|
61 |
+
Dimension of the text context vector (default: 512).
|
62 |
+
use_meanflow (bool):
|
63 |
+
If True, uses the MeanFlow generation method; otherwise, uses FlowMatching. If variant is 'meanaudio_mf', this is automatically set to True (default: False).
|
64 |
+
'''
|
65 |
+
setup_eval_logging()
|
66 |
+
output_dir = Path(output).expanduser()
|
67 |
+
output_dir.mkdir(parents=True, exist_ok=True)
|
68 |
+
device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
69 |
+
dtype = torch.float32 if full_precision else torch.bfloat16
|
70 |
+
if duration <= 0 or num_steps <= 0:
|
71 |
+
raise ValueError("Duration and number of steps must be positive.")
|
72 |
+
if variant not in all_model_cfg:
|
73 |
+
raise ValueError(f"Unknown model variant: {variant}. Available: {list(all_model_cfg.keys())}")
|
74 |
+
if not model_path or model_path == '':
|
75 |
+
model_path = Path(f'./weights/{variant}.pth')
|
76 |
+
else:
|
77 |
+
model_path = Path(model_path)
|
78 |
+
if not model_path.exists():
|
79 |
+
if str(model_path) == f'./weights/{variant}.pth':
|
80 |
+
log.info(f'Model not found at {model_path}')
|
81 |
+
log.info('Downloading models to "./weights/"...')
|
82 |
+
try:
|
83 |
+
weights_dir = Path('./weights')
|
84 |
+
weights_dir.mkdir(exist_ok=True)
|
85 |
+
snapshot_download(repo_id="junxiliu/Meanaudio", local_dir="./weights",allow_patterns=["*.pt", "*.pth"] )
|
86 |
+
raise NotImplementedError("Model download functionality needs to be implemented")
|
87 |
+
except Exception as e:
|
88 |
+
log.error(f"Failed to download model: {e}")
|
89 |
+
raise FileNotFoundError(f"Model file not found and download failed: {model_path}")
|
90 |
+
else:
|
91 |
+
raise FileNotFoundError(f"Model file not found: {model_path}")
|
92 |
+
|
93 |
+
model = all_model_cfg[variant]
|
94 |
+
seq_cfg = model.seq_cfg
|
95 |
+
seq_cfg.duration = duration
|
96 |
+
|
97 |
+
net = get_mean_audio(model.model_name, use_rope=use_rope, text_c_dim=text_c_dim)
|
98 |
+
net = net.to(device, dtype).eval()
|
99 |
+
net.load_weights(torch.load(model_path, map_location=device, weights_only=True))
|
100 |
+
net.update_seq_lengths(seq_cfg.latent_seq_len)
|
101 |
+
|
102 |
+
if variant=='meanaudio_mf':
|
103 |
+
use_meanflow=True
|
104 |
+
if use_meanflow:
|
105 |
+
generation_func = MeanFlow(steps=num_steps)
|
106 |
+
cfg_strength=0
|
107 |
+
else:
|
108 |
+
generation_func = FlowMatching(min_sigma=0, inference_mode='euler', num_steps=num_steps)
|
109 |
+
|
110 |
+
feature_utils = FeaturesUtils(
|
111 |
+
tod_vae_ckpt=model.vae_path,
|
112 |
+
enable_conditions=True,
|
113 |
+
encoder_name=encoder_name,
|
114 |
+
mode=model.mode,
|
115 |
+
bigvgan_vocoder_ckpt=model.bigvgan_16k_path,
|
116 |
+
need_vae_encoder=False
|
117 |
+
)
|
118 |
+
feature_utils = feature_utils.to(device, dtype).eval()
|
119 |
+
|
120 |
+
rng = torch.Generator(device=device)
|
121 |
+
rng.manual_seed(seed)
|
122 |
+
|
123 |
+
generate_fn = generate_mf if use_meanflow else generate_fm
|
124 |
+
kwargs = {
|
125 |
+
'negative_text': [negative_prompt],
|
126 |
+
'feature_utils': feature_utils,
|
127 |
+
'net': net,
|
128 |
+
'rng': rng,
|
129 |
+
'cfg_strength': cfg_strength
|
130 |
+
}
|
131 |
+
|
132 |
+
if use_meanflow:
|
133 |
+
kwargs['mf'] = generation_func
|
134 |
+
else:
|
135 |
+
kwargs['fm'] = generation_func
|
136 |
+
|
137 |
+
audios = generate_fn([prompt], **kwargs)
|
138 |
+
audio = audios.float().cpu()[0]
|
139 |
+
safe_filename = prompt.replace(' ', '_').replace('/', '_').replace('.', '')
|
140 |
+
save_path = output_dir / f'{safe_filename}--numsteps{num_steps}--seed{seed}.wav'
|
141 |
+
torchaudio.save(save_path, audio, seq_cfg.sampling_rate)
|
142 |
+
log.info(f'Audio saved to {save_path}')
|
143 |
+
log.info('Memory usage: %.2f GB', torch.cuda.max_memory_allocated() / (2**30))
|
144 |
+
return save_path
|
145 |
+
|
146 |
+
if __name__ == '__main__':
|
147 |
+
MeanAudioInference('a dog is barking')
|
app.py
ADDED
@@ -0,0 +1,421 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
import warnings
|
3 |
+
|
4 |
+
warnings.filterwarnings("ignore", category=FutureWarning)
|
5 |
+
import logging
|
6 |
+
from argparse import ArgumentParser
|
7 |
+
from pathlib import Path
|
8 |
+
import torch
|
9 |
+
import torchaudio
|
10 |
+
import gradio as gr
|
11 |
+
from meanaudio.eval_utils import (
|
12 |
+
ModelConfig,
|
13 |
+
all_model_cfg,
|
14 |
+
generate_mf,
|
15 |
+
generate_fm,
|
16 |
+
setup_eval_logging,
|
17 |
+
)
|
18 |
+
from meanaudio.model.flow_matching import FlowMatching
|
19 |
+
from meanaudio.model.mean_flow import MeanFlow
|
20 |
+
from meanaudio.model.networks import MeanAudio, get_mean_audio
|
21 |
+
from meanaudio.model.utils.features_utils import FeaturesUtils
|
22 |
+
|
23 |
+
torch.backends.cuda.matmul.allow_tf32 = True
|
24 |
+
torch.backends.cudnn.allow_tf32 = True
|
25 |
+
import gc
|
26 |
+
from datetime import datetime
|
27 |
+
|
28 |
+
log = logging.getLogger()
|
29 |
+
|
30 |
+
device = "cpu"
|
31 |
+
if torch.cuda.is_available():
|
32 |
+
device = "cuda"
|
33 |
+
elif torch.backends.mps.is_available():
|
34 |
+
device = "mps"
|
35 |
+
else:
|
36 |
+
log.warning("CUDA/MPS are not available, running on CPU")
|
37 |
+
setup_eval_logging()
|
38 |
+
|
39 |
+
OUTPUT_DIR = Path("./output/gradio")
|
40 |
+
OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
|
41 |
+
|
42 |
+
current_model_state = {
|
43 |
+
"net": None,
|
44 |
+
"feature_utils": None,
|
45 |
+
"seq_cfg": None,
|
46 |
+
"args": None,
|
47 |
+
}
|
48 |
+
|
49 |
+
|
50 |
+
def load_model_if_needed(
|
51 |
+
variant, model_path, encoder_name, use_rope, text_c_dim, full_precision
|
52 |
+
):
|
53 |
+
global current_model_state
|
54 |
+
dtype = torch.float32 if full_precision else torch.bfloat16
|
55 |
+
needs_reload = (
|
56 |
+
current_model_state["args"] is None
|
57 |
+
or current_model_state["args"].variant != variant
|
58 |
+
or current_model_state["args"].model_path != model_path
|
59 |
+
or current_model_state["args"].encoder_name != encoder_name
|
60 |
+
or current_model_state["args"].use_rope != use_rope
|
61 |
+
or current_model_state["args"].text_c_dim != text_c_dim
|
62 |
+
or current_model_state["args"].full_precision != full_precision
|
63 |
+
)
|
64 |
+
if needs_reload:
|
65 |
+
try:
|
66 |
+
if variant not in all_model_cfg:
|
67 |
+
raise ValueError(f"Unknown model variant: {variant}")
|
68 |
+
model: ModelConfig = all_model_cfg[variant]
|
69 |
+
seq_cfg = model.seq_cfg
|
70 |
+
|
71 |
+
class MockArgs:
|
72 |
+
pass
|
73 |
+
|
74 |
+
mock_args = MockArgs()
|
75 |
+
mock_args.variant = variant
|
76 |
+
mock_args.model_path = model_path
|
77 |
+
mock_args.encoder_name = encoder_name
|
78 |
+
mock_args.use_rope = use_rope
|
79 |
+
mock_args.text_c_dim = text_c_dim
|
80 |
+
mock_args.full_precision = full_precision
|
81 |
+
|
82 |
+
net: MeanAudio = (
|
83 |
+
get_mean_audio(
|
84 |
+
model.model_name,
|
85 |
+
use_rope=mock_args.use_rope,
|
86 |
+
text_c_dim=mock_args.text_c_dim,
|
87 |
+
)
|
88 |
+
.to(device, dtype)
|
89 |
+
.eval()
|
90 |
+
)
|
91 |
+
net.load_weights(
|
92 |
+
torch.load(
|
93 |
+
mock_args.model_path, map_location=device, weights_only=True
|
94 |
+
)
|
95 |
+
)
|
96 |
+
log.info(f"Loaded weights from {mock_args.model_path}")
|
97 |
+
|
98 |
+
feature_utils = FeaturesUtils(
|
99 |
+
tod_vae_ckpt=model.vae_path,
|
100 |
+
enable_conditions=True,
|
101 |
+
encoder_name=mock_args.encoder_name,
|
102 |
+
mode=model.mode,
|
103 |
+
bigvgan_vocoder_ckpt=model.bigvgan_16k_path,
|
104 |
+
need_vae_encoder=False,
|
105 |
+
)
|
106 |
+
feature_utils = feature_utils.to(device, dtype).eval()
|
107 |
+
|
108 |
+
current_model_state["net"] = net
|
109 |
+
current_model_state["feature_utils"] = feature_utils
|
110 |
+
current_model_state["seq_cfg"] = seq_cfg
|
111 |
+
current_model_state["args"] = mock_args
|
112 |
+
log.info(f"Model '{variant}' loaded successfully.")
|
113 |
+
return True
|
114 |
+
except Exception as e:
|
115 |
+
log.error(f"Error loading model: {e}")
|
116 |
+
|
117 |
+
current_model_state = {
|
118 |
+
"net": None,
|
119 |
+
"feature_utils": None,
|
120 |
+
"seq_cfg": None,
|
121 |
+
"args": None,
|
122 |
+
}
|
123 |
+
raise e
|
124 |
+
else:
|
125 |
+
log.info(f"Model '{variant}' already loaded with current settings.")
|
126 |
+
return False
|
127 |
+
|
128 |
+
|
129 |
+
@torch.inference_mode()
|
130 |
+
def generate_audio_gradio(
|
131 |
+
prompt,
|
132 |
+
negative_prompt,
|
133 |
+
duration,
|
134 |
+
cfg_strength,
|
135 |
+
num_steps,
|
136 |
+
seed,
|
137 |
+
variant,
|
138 |
+
full_precision,
|
139 |
+
):
|
140 |
+
global current_model_state
|
141 |
+
use_meanflow = variant == "meanaudio_mf"
|
142 |
+
|
143 |
+
model_path = (
|
144 |
+
"./weights/meanaudio_mf.pth"
|
145 |
+
if use_meanflow
|
146 |
+
else "./weights/fluxaudio_fm.pth"
|
147 |
+
)
|
148 |
+
encoder_name = "t5_clap"
|
149 |
+
use_rope = True
|
150 |
+
text_c_dim = 512
|
151 |
+
|
152 |
+
try:
|
153 |
+
load_model_if_needed(
|
154 |
+
variant, model_path, encoder_name, use_rope, text_c_dim, full_precision
|
155 |
+
)
|
156 |
+
except Exception as e:
|
157 |
+
return f"Error loading model: {str(e)}", None
|
158 |
+
|
159 |
+
if current_model_state["net"] is None:
|
160 |
+
return "Error: Model could not be loaded.", None
|
161 |
+
net = current_model_state["net"]
|
162 |
+
feature_utils = current_model_state["feature_utils"]
|
163 |
+
seq_cfg = current_model_state["seq_cfg"]
|
164 |
+
|
165 |
+
args = current_model_state["args"]
|
166 |
+
dtype = torch.float32 if args.full_precision else torch.bfloat16
|
167 |
+
|
168 |
+
try:
|
169 |
+
seq_cfg.duration = duration
|
170 |
+
net.update_seq_lengths(seq_cfg.latent_seq_len)
|
171 |
+
|
172 |
+
rng = torch.Generator(device=device)
|
173 |
+
if seed >= 0:
|
174 |
+
rng.manual_seed(seed)
|
175 |
+
else:
|
176 |
+
rng.seed()
|
177 |
+
|
178 |
+
if use_meanflow:
|
179 |
+
sampler = MeanFlow(steps=num_steps)
|
180 |
+
log.info("Using MeanFlow for generation.")
|
181 |
+
generation_func = generate_mf
|
182 |
+
sampler_arg_name = "mf"
|
183 |
+
cfg_strength = 3
|
184 |
+
else:
|
185 |
+
sampler = FlowMatching(
|
186 |
+
min_sigma=0, inference_mode="euler", num_steps=num_steps
|
187 |
+
)
|
188 |
+
log.info("Using FlowMatching for generation.")
|
189 |
+
generation_func = generate_fm
|
190 |
+
sampler_arg_name = "fm"
|
191 |
+
|
192 |
+
prompts = [prompt]
|
193 |
+
|
194 |
+
audios = generation_func(
|
195 |
+
prompts,
|
196 |
+
negative_text=[negative_prompt],
|
197 |
+
feature_utils=feature_utils,
|
198 |
+
net=net,
|
199 |
+
rng=rng,
|
200 |
+
cfg_strength=cfg_strength,
|
201 |
+
**{sampler_arg_name: sampler},
|
202 |
+
)
|
203 |
+
audio = audios.float().cpu()[0]
|
204 |
+
|
205 |
+
safe_prompt = (
|
206 |
+
"".join(c for c in prompt if c.isalnum() or c in (" ", "_"))
|
207 |
+
.rstrip()
|
208 |
+
.replace(" ", "_")[:50]
|
209 |
+
)
|
210 |
+
current_time_string = datetime.now().strftime("%Y%m%d_%H%M%S_%f")
|
211 |
+
filename = f"{safe_prompt}_{current_time_string}.flac"
|
212 |
+
save_path = OUTPUT_DIR / filename
|
213 |
+
torchaudio.save(str(save_path), audio, seq_cfg.sampling_rate)
|
214 |
+
log.info(f"Audio saved to {save_path}")
|
215 |
+
|
216 |
+
gc.collect()
|
217 |
+
|
218 |
+
return (
|
219 |
+
f"Generated audio for prompt: '{prompt}' using {'MeanFlow' if use_meanflow else 'FlowMatching'}",
|
220 |
+
str(save_path),
|
221 |
+
)
|
222 |
+
except Exception as e:
|
223 |
+
gc.collect()
|
224 |
+
log.error(f"Generation error: {e}")
|
225 |
+
return f"Error during generation: {str(e)}", None
|
226 |
+
|
227 |
+
|
228 |
+
theme = gr.themes.Soft(
|
229 |
+
primary_hue="blue",
|
230 |
+
secondary_hue="slate",
|
231 |
+
neutral_hue="slate",
|
232 |
+
text_size="sm",
|
233 |
+
spacing_size="sm",
|
234 |
+
).set(
|
235 |
+
background_fill_primary="*neutral_50",
|
236 |
+
background_fill_secondary="*background_fill_primary",
|
237 |
+
block_background_fill="*background_fill_primary",
|
238 |
+
block_border_width="0px",
|
239 |
+
panel_background_fill="*neutral_50",
|
240 |
+
panel_border_width="0px",
|
241 |
+
input_background_fill="*neutral_100",
|
242 |
+
input_border_color="*neutral_200",
|
243 |
+
button_primary_background_fill="*primary_300",
|
244 |
+
button_primary_background_fill_hover="*primary_400",
|
245 |
+
button_secondary_background_fill="*neutral_200",
|
246 |
+
button_secondary_background_fill_hover="*neutral_300",
|
247 |
+
)
|
248 |
+
|
249 |
+
custom_css = """
|
250 |
+
#main-header {
|
251 |
+
text-align: center;
|
252 |
+
margin-top: 5px;
|
253 |
+
margin-bottom: 10px;
|
254 |
+
color: var(--neutral-600);
|
255 |
+
font-weight: 600;
|
256 |
+
}
|
257 |
+
#model-settings-header, #generation-settings-header {
|
258 |
+
color: var(--neutral-600);
|
259 |
+
margin-top: 8px;
|
260 |
+
margin-bottom: 8px;
|
261 |
+
font-weight: 500;
|
262 |
+
font-size: 1.1em;
|
263 |
+
}
|
264 |
+
.setting-section {
|
265 |
+
padding: 10px 12px;
|
266 |
+
border-radius: 6px;
|
267 |
+
background-color: var(--neutral-50);
|
268 |
+
margin-bottom: 10px;
|
269 |
+
border: 1px solid var(--neutral-100);
|
270 |
+
}
|
271 |
+
hr {
|
272 |
+
border: none;
|
273 |
+
height: 1px;
|
274 |
+
background-color: var(--neutral-200);
|
275 |
+
margin: 8px 0;
|
276 |
+
}
|
277 |
+
#generate-btn {
|
278 |
+
width: 100%;
|
279 |
+
max-width: 250px;
|
280 |
+
margin: 10px auto;
|
281 |
+
display: block;
|
282 |
+
padding: 10px 15px;
|
283 |
+
font-size: 16px;
|
284 |
+
border-radius: 5px;
|
285 |
+
}
|
286 |
+
#status-box {
|
287 |
+
min-height: 50px;
|
288 |
+
display: flex;
|
289 |
+
align-items: center;
|
290 |
+
justify-content: center;
|
291 |
+
padding: 8px;
|
292 |
+
border-radius: 5px;
|
293 |
+
border: 1px solid var(--neutral-200);
|
294 |
+
color: var(--neutral-700);
|
295 |
+
}
|
296 |
+
#audio-output {
|
297 |
+
height: 100px;
|
298 |
+
border-radius: 5px;
|
299 |
+
border: 1px solid var(--neutral-200);
|
300 |
+
}
|
301 |
+
.gradio-dropdown label, .gradio-checkbox label, .gradio-number label, .gradio-textbox label {
|
302 |
+
font-weight: 500;
|
303 |
+
color: var(--neutral-700);
|
304 |
+
font-size: 0.9em;
|
305 |
+
}
|
306 |
+
.gradio-row {
|
307 |
+
gap: 8px;
|
308 |
+
}
|
309 |
+
.gradio-block {
|
310 |
+
margin-bottom: 8px;
|
311 |
+
}
|
312 |
+
.setting-section .gradio-block {
|
313 |
+
margin-bottom: 6px;
|
314 |
+
}
|
315 |
+
::-webkit-scrollbar {
|
316 |
+
width: 8px;
|
317 |
+
height: 8px;
|
318 |
+
}
|
319 |
+
::-webkit-scrollbar-track {
|
320 |
+
background: var(--neutral-100);
|
321 |
+
border-radius: 4px;
|
322 |
+
}
|
323 |
+
::-webkit-scrollbar-thumb {
|
324 |
+
background: var(--neutral-300);
|
325 |
+
border-radius: 4px;
|
326 |
+
}
|
327 |
+
::-webkit-scrollbar-thumb:hover {
|
328 |
+
background: var(--neutral-400);
|
329 |
+
}
|
330 |
+
* {
|
331 |
+
scrollbar-width: thin;
|
332 |
+
scrollbar-color: var(--neutral-300) var(--neutral-100);
|
333 |
+
}
|
334 |
+
"""
|
335 |
+
|
336 |
+
with gr.Blocks(title="MeanAudio Generator", theme=theme, css=custom_css) as demo:
|
337 |
+
gr.Markdown("# MeanAudio Text-to-Audio Generator", elem_id="main-header")
|
338 |
+
|
339 |
+
gr.Markdown("### Model and Generation Settings", elem_id="model-settings-header")
|
340 |
+
with gr.Column(elem_classes="setting-section"):
|
341 |
+
with gr.Row():
|
342 |
+
available_variants = (
|
343 |
+
list(all_model_cfg.keys()) if all_model_cfg else []
|
344 |
+
)
|
345 |
+
default_variant = (
|
346 |
+
"small_16k_mf"
|
347 |
+
if "small_16k_mf" in available_variants
|
348 |
+
else available_variants[0] if available_variants else ""
|
349 |
+
)
|
350 |
+
variant = gr.Dropdown(
|
351 |
+
label="Model Variant",
|
352 |
+
choices=available_variants,
|
353 |
+
value=default_variant,
|
354 |
+
interactive=True,
|
355 |
+
scale=3,
|
356 |
+
)
|
357 |
+
full_precision = gr.Checkbox(
|
358 |
+
label="Full Precision (float32)", value=True, scale=1
|
359 |
+
)
|
360 |
+
|
361 |
+
gr.Markdown("### Audio Generation", elem_id="generation-settings-header")
|
362 |
+
with gr.Column(elem_classes="setting-section"):
|
363 |
+
with gr.Row():
|
364 |
+
prompt = gr.Textbox(
|
365 |
+
label="Prompt",
|
366 |
+
placeholder="Describe the sound you want to generate...",
|
367 |
+
scale=1,
|
368 |
+
)
|
369 |
+
negative_prompt = gr.Textbox(
|
370 |
+
label="Negative Prompt",
|
371 |
+
placeholder="Describe sounds you want to avoid...",
|
372 |
+
value="",
|
373 |
+
scale=1,
|
374 |
+
)
|
375 |
+
with gr.Row():
|
376 |
+
duration = gr.Number(
|
377 |
+
label="Duration (sec)", value=10.0, minimum=0.1, scale=1
|
378 |
+
)
|
379 |
+
cfg_strength = gr.Number(
|
380 |
+
label="CFG (Meanflow forced to 3)", value=3, minimum=0.0, scale=1
|
381 |
+
)
|
382 |
+
with gr.Row():
|
383 |
+
seed = gr.Number(
|
384 |
+
label="Seed (-1 for random)", value=42, precision=0, scale=1
|
385 |
+
)
|
386 |
+
num_steps = gr.Number(
|
387 |
+
label="Number of Steps",
|
388 |
+
value=1,
|
389 |
+
precision=0,
|
390 |
+
minimum=1,
|
391 |
+
scale=1,
|
392 |
+
)
|
393 |
+
|
394 |
+
generate_button = gr.Button("Generate", variant="primary", elem_id="generate-btn")
|
395 |
+
generate_output_text = gr.Textbox(
|
396 |
+
label="Result Status", interactive=False, elem_id="status-box"
|
397 |
+
)
|
398 |
+
audio_output = gr.Audio(
|
399 |
+
label="Generated Audio", type="filepath", elem_id="audio-output"
|
400 |
+
)
|
401 |
+
generate_button.click(
|
402 |
+
fn=generate_audio_gradio,
|
403 |
+
inputs=[
|
404 |
+
prompt,
|
405 |
+
negative_prompt,
|
406 |
+
duration,
|
407 |
+
cfg_strength,
|
408 |
+
num_steps,
|
409 |
+
seed,
|
410 |
+
variant,
|
411 |
+
full_precision,
|
412 |
+
],
|
413 |
+
outputs=[generate_output_text, audio_output],
|
414 |
+
)
|
415 |
+
|
416 |
+
if __name__ == "__main__":
|
417 |
+
parser = ArgumentParser()
|
418 |
+
parser.add_argument("--port", type=int, default=7861)
|
419 |
+
args = parser.parse_args()
|
420 |
+
demo.launch(server_port=args.port, allowed_paths=[OUTPUT_DIR.resolve()])
|
421 |
+
|
easyinfer.py
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
from MeanAudio import MeanAudioInference
|
2 |
+
audio_path=MeanAudioInference('a dog is barking')
|
3 |
+
print(audio_path)
|
requirements.txt
ADDED
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
torch>=2.5.1
|
2 |
+
huggingface_hub>=0.26
|
3 |
+
cython
|
4 |
+
gitpython>=3.1
|
5 |
+
tensorboard>=2.11
|
6 |
+
numpy>=1.21,<2.1
|
7 |
+
Pillow>=9.5
|
8 |
+
opencv-python>=4.8
|
9 |
+
scipy>=1.7
|
10 |
+
tqdm>=4.66.1
|
11 |
+
gradio>=3.34
|
12 |
+
einops>=0.6
|
13 |
+
hydra-core>=1.3.2
|
14 |
+
requests
|
15 |
+
torchdiffeq>=0.2.5
|
16 |
+
librosa>=0.8.1
|
17 |
+
nitrous-ema
|
18 |
+
hydra_colorlog
|
19 |
+
tensordict>=0.6.1
|
20 |
+
colorlog
|
21 |
+
open_clip_torch>=2.29.0
|
22 |
+
av>=14.0.1
|
23 |
+
timm>=1.0.12
|
24 |
+
python-dotenv
|
25 |
+
transformers
|
26 |
+
debugpy
|
27 |
+
laion-clap
|