run

app.py

@@ -22,7 +22,7 @@ def transcribe_and_respond(audio_file):
         print(f"Audio dtype: {audio.dtype}, Audio shape: {audio.shape}, Sample rate: {sr}")
 
         turns = [
-            {'role': 'system', 'content': 'Respond naturally and informatively.'},
+            {'role': 'system', 'content': 'You are a transcription agent. What ever the speaker has spoken should be outputted as text. Nothing less, nothing more.'},
             {'role': 'user', 'content': '<|audio|>'}
         ]
 
@@ -30,7 +30,7 @@ def transcribe_and_respond(audio_file):
         print(f"Initial turns: {turns}")
 
         # Call the model with the audio and prompt
-        output = pipe({'audio': audio, 'turns': turns, 'sampling_rate': sr}, max_new_tokens=512)
+        output = pipe({'audio': audio, 'turns': turns, 'sampling_rate': sr}, max_new_tokens=1024)
 
         # Debug: Print the final output from the model
         print(f"Model output: {output}")