Update app.py

app.py

@@ -138,8 +138,10 @@ def predict(text, speaker):
     ### ### ### 
     example = dataset['test'][11]
     # speaker_embedding = torch.tensor(example["speaker_embeddings"]).unsqueeze(0)
+    speaker_embedding = torch.tensor(example["speaker_embeddings"]).unsqueeze(0).unsqueeze(0).to(device)
     speaker_embedding = speaker_embedding.expand(-1, inputs["input_ids"].size(1), -1)
     spectrogram = model.generate_speech(inputs["input_ids"].to(device), speaker_embedding)
+    
     # speaker_embedding = torch.tensor(speaker_embedding).unsqueeze(0)
     # spectrogram = model.generate_speech(inputs["input_ids"], speaker_embedding)
     with torch.no_grad():