Update barks.py

barks.py

@@ -283,12 +283,24 @@ def generate_vocals(vocal_prompt: str, total_duration: int):
     
     try:
         print("Generating vocals with Bark...")
-        # Process vocal prompt, ensuring inputs are on cuda
-        inputs = bark_processor(vocal_prompt, return_tensors="pt").to("cuda")
+        # Process vocal prompt with explicit padding and attention mask
+        inputs = bark_processor(
+            vocal_prompt,
+            return_tensors="pt",
+            padding=True,
+            return_attention_mask=True
+        ).to("cuda")
         
+        # Set pad_token_id explicitly, avoiding eos_token_id
+        pad_token_id = 0  # Use 0 as a safe padding token
         # Generate vocals with mixed precision
         with torch.no_grad(), autocast():
-            vocal_array = bark_model.generate(**inputs, do_sample=True)
+            vocal_array = bark_model.generate(
+                input_ids=inputs["input_ids"],
+                attention_mask=inputs["attention_mask"],
+                do_sample=True,
+                pad_token_id=pad_token_id
+            )
         
         # Convert to numpy and create AudioSegment
         vocal_array = vocal_array.cpu().numpy().squeeze()