11

app.py

@@ -18,10 +18,9 @@ def text_to_speech(text, processor, model, vocoder):
         # Process the input text
         inputs = processor(text=text, return_tensors="pt")
         
-        # Load speaker embeddings (use a default speaker)
-        from datasets import load_dataset
-        embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
-        speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
+        # Create a simple default speaker embedding (zeros vector)
+        # This is a fallback when specific speaker embeddings are not available
+        speaker_embeddings = torch.zeros((1, 512))  # Standard speaker embedding size
         
         # Generate speech using the correct method
         with torch.no_grad():