Update app.py

app.py

@@ -65,20 +65,6 @@ def combine_audio_files(audio_files):
         os.remove(audio_file)  # Remove temporary files
     return combined
 
-def postprocess(prediction_value):
-    if isinstance(prediction_value, pydub.AudioSegment):
-        # Convert AudioSegment to numpy array
-        samples = np.array(prediction_value.get_array_of_samples())
-        # If stereo, take the mean of both channels
-        if prediction_value.channels == 2:
-            samples = np.mean(samples.reshape(-1, 2), axis=1)
-        # Return as tuple (sample_rate, samples)
-        return (prediction_value.frame_rate, samples.astype(np.float32))
-    else:
-        raise ValueError(f"Cannot process {type(prediction_value)} as Audio")
-
-    
-
 #guardrail model
 guard_llm = "llama-3.1-8b-instant"
 
@@ -138,7 +124,7 @@ async def greet(product,description):
         audio_file = text_to_speech(a_list[0])
         audio_files.append(audio_file)
         final_audio = combine_audio_files(audio_files)
-        yield (final_audio.frame_rate, np.array(final_audio.get_array_of_samples()))
+        yield (final_audio.frame_rate, final_audio)
     else:
         audio_files = []
         output = llm.create_chat_completion(
@@ -160,7 +146,7 @@ async def greet(product,description):
                 audio_file = text_to_speech(delta.get('content', ''))
                 audio_files.append(audio_file)
         final_audio = combine_audio_files(audio_files)
-        yield (final_audio.frame_rate, np.array(final_audio.get_array_of_samples()))
+        yield (final_audio.frame_rate, final_audio)
 
 demo = gr.Interface(fn=greet, inputs=["text","text"], outputs=gr.Audio(), concurrency_limit=10)
 demo.launch()