Update app.py

app.py

@@ -130,18 +130,20 @@ def predict(text, speaker):
         return (16000, np.zeros(0).astype(np.int16))
 
     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]
 
     ### ### ### 
     example = dataset['test'][11]
     speaker_embedding = torch.tensor(example["speaker_embeddings"]).unsqueeze(0)
-
+    
     speaker_embedding = torch.tensor(speaker_embedding).unsqueeze(0)
-
-    speech = model.generate_speech(input_ids, speaker_embedding, vocoder=vocoder)
+    spectrogram = model.generate_speech(inputs["input_ids"], speaker_embedding)
+    with torch.no_grad():
+        speech = vocoder(spectrogram)
+    # speech = model.generate_speech(input_ids, speaker_embedding, vocoder=vocoder)
 
     speech = (speech.numpy() * 32767).astype(np.int16)
     return (16000, speech)