Update app.py

app.py

@@ -48,7 +48,8 @@ def predict(temp_text, temp_audio, text):
     with torch.no_grad():
         speech = vocoder(spectrogram)
  
-    return speech.numpy()
+    speech = (speech.numpy() * 32767).astype(np.int16)
+    return (16000, speech)