Upload app.py

app.py

@@ -3,6 +3,7 @@ import torch
 import soundfile as sf
 from speechbrain.inference.TTS import Tacotron2
 from speechbrain.inference.vocoders import HIFIGAN
+from speechbrain.dataio.dataio import read_audio
 
 # モデルのロード
 hifi_gan = HIFIGAN.from_hparams(source="speechbrain/tts-hifigan-ljspeech", savedir="tmpdir_vocoder")
@@ -11,7 +12,8 @@ tacotron2 = Tacotron2.from_hparams(source="speechbrain/tts-tacotron2-ljspeech",
 # 推論関数の定義
 def synthesize_speech(text):
     # テキストをトークンIDに変換
-    tokenized = tacotron2.tokenizer(text, phonemize=False)
+    text = text.lower()
+    tokenized = tacotron2.hparams.tokenize(text, phonemize=False)
     # トークンIDをLong型のテンソルに変換
     tokens = torch.LongTensor(tokenized)
 
@@ -34,45 +36,4 @@ iface = gr.Interface(
     description="Enter text to synthesize speech."
 )
 
-iface.launch()
-
-# import gradio as gr
-# import torch
-# from speechbrain.inference.TTS import Tacotron2
-# from speechbrain.inference.vocoders import HIFIGAN
-
-# # モデルのロード
-# hifi_gan = HIFIGAN.from_hparams(source="speechbrain/tts-hifigan-ljspeech", savedir="tmpdir_vocoder")
-# tacotron2 = Tacotron2.from_hparams(source="speechbrain/tts-tacotron2-ljspeech", savedir="tmpdir_tts")
-
-# # 推論関数の定義
-# def synthesize_speech(text):
-#     # Tacotron2でmel spectrogramを生成
-#     # テキストを直接入力として、LongTensorでラップする
-#     mel_output, mel_length, alignment = tacotron2.encode_batch([text])
-
-#     # HiFi-GANでmel spectrogramから音声を生成
-#     waveforms = hifi_gan.decode_batch(mel_output)
-
-#     # torch tensorをwavfileとして保存
-#     torch.save(waveforms, "speech.pt")
-#     return "speech.pt"
-
-# # Gradioインターフェースの作成
-# iface = gr.Interface(
-#     fn=synthesize_speech,
-#     inputs=gr.Textbox(lines=5, label="Input Text"),
-#     outputs=gr.Audio(label="Output Audio", type="filepath"),
-#     title="TTS Demo",
-#     description="Enter text to synthesize speech."
-# )
-
-# iface.launch()
-
-
-
-
-
-
-
-
+iface.launch()