Update app.py

app.py

@@ -28,8 +28,8 @@ def predict(text, speaker):
     inputs = processor(text=text, return_tensors="pt")
 
     # limit input length
-    #input_ids = inputs["input_ids"]
-    #input_ids = input_ids[..., :model.config.max_text_positions]
+    input_ids = inputs["input_ids"]
+    input_ids = input_ids[..., :model.config.max_text_positions]
     
     if speaker == "Surprise Me!":
         # load one of the provided speaker embeddings at random
@@ -51,8 +51,8 @@ def predict(text, speaker):
 
     #speaker_embedding = torch.tensor(speaker_embedding).unsqueeze(0)
     speaker_embedding = torch.tensor(speaker_embedding) #the saved model is already unsqueezed, but is not a tensor, so make it one
-  #  speech = model.generate_speech(input_ids, speaker_embedding, vocoder=vocoder)
-    speech = model.generate_speech(inputs["input_ids"], speaker_embedding, vocoder=vocoder)
+    speech = model.generate_speech(input_ids, speaker_embedding, vocoder=vocoder)
+    #speech = model.generate_speech(inputs["input_ids"], speaker_embedding, vocoder=vocoder)
 
     speech = (speech.numpy() * 32767).astype(np.int16)
     return (16000, speech)