Extract only the assistant's response from the generated text

transcript.py

@@ -114,7 +114,12 @@ def main():
 
         # Inference: Generation of the output text
         text_ids, _ = model.generate(**inputs, use_audio_in_video=False)
-        text = processor.batch_decode(text_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        full_text = processor.batch_decode(text_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        # Extract only the assistant's response
+        if "assistant\n" in full_text:
+            text = full_text.split("assistant\n")[-1].strip()
+        else:
+            text = full_text.strip()
 
         # Format timestamps
         start_min, start_sec = divmod(start, 60)