Update app.py

app.py

@@ -50,7 +50,7 @@ def predict(text, speaker):
         speaker_embedding = np.load(speaker_embeddings[speaker[:3]])
 
     #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 = (speech.numpy() * 32767).astype(np.int16)