Update app.py

app.py

@@ -111,8 +111,11 @@ def transcribe_audio(audio_file, chunk_length_s=30):
         # Create attention mask
         attention_mask = torch.ones(inputs.input_features.shape[:2], dtype=torch.long, device=device)
 
+        # input_ids used to set attention_mask to zero for padding tokens
+        input_ids = inputs['input_ids']  
+
         # Set the attention mask to zero for padding tokens
-        attention_mask[input_features == processor.tokenizer.pad_token_id] = 0
+        attention_mask[input_ids == processor.tokenizer.pad_token_id] = 0
 
 
         # ASR model inference on the chunk