Upload app.py

app.py

@@ -1,8 +1,9 @@
 import gradio as gr
 import torch
+import soundfile as sf
 from speechbrain.inference.TTS import Tacotron2
 from speechbrain.inference.vocoders import HIFIGAN
-import soundfile as sf
+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")
@@ -10,17 +11,18 @@ 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
+    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)
+
     # 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])
+    mel_output, mel_length, alignment = tacotron2.encode_batch(tokens)
 
     # 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"
-
     # 音声を .wav 形式で保存
     sf.write("speech.wav", waveforms.squeeze().cpu().numpy(), samplerate=hifi_gan.hparams.sample_rate)
     return "speech.wav"
@@ -34,12 +36,4 @@ iface = gr.Interface(
     description="Enter text to synthesize speech."
 )
 
-iface.launch()
-
-
-
-
-
-
-
-
+iface.launch()