Upload 2 files

app.py

@@ -2,6 +2,7 @@ import gradio as gr
 import torch
 from speechbrain.inference.TTS import Tacotron2
 from speechbrain.inference.vocoders import HIFIGAN
+from scipy.io.wavfile import write
 
 # モデルのロード
 hifi_gan = HIFIGAN.from_hparams(source="speechbrain/tts-hifigan-ljspeech", savedir="tmpdir_vocoder")
@@ -17,8 +18,9 @@ def synthesize_speech(text):
     waveforms = hifi_gan.decode_batch(mel_output)
 
     # torch tensorをwavfileとして保存
-    torch.save(waveforms, "speech.pt")
-    return "speech.pt"
+    # sampling rate を 22050に設定
+    write("speech.wav", rate=22050, data=waveforms.squeeze().cpu().numpy())
+    return "speech.wav"
 
 # Gradioインターフェースの作成
 iface = gr.Interface(