Update app.py

app.py

@@ -97,7 +97,6 @@ def transcribe_function(stream, new_chunk):
         print(f"Error chunk structure: {type(new_chunk)}, content: {new_chunk}")
         return stream, "", None
 
-    # Ensure y is not empty and is at least 1-dimensional
     if y is None or len(y) == 0:
         return stream, "", None
 
@@ -106,12 +105,22 @@ def transcribe_function(stream, new_chunk):
     if max_abs_y > 0:
         y = y / max_abs_y
 
-    # Ensure stream is also at least 1-dimensional before concatenation
     if stream is not None and len(stream) > 0:
         stream = np.concatenate([stream, y])
     else:
         stream = y
 
+    # Only process if there is significant audio input
+    if len(stream) > sr * 0.5:  # Process if there is at least 0.5 seconds of audio
+        result = pipe_asr({"array": stream, "sampling_rate": sr}, return_timestamps=False)
+        full_text = result.get("text", "")
+        if full_text.strip():
+            print(f"Transcription: {full_text}")
+            threading.Thread(target=auto_reset_state).start()
+            return stream, full_text, full_text
+
+    return stream, "", None
+
     # Process the audio data for transcription
     result = pipe_asr({"array": stream, "sampling_rate": sr}, return_timestamps=False)
     full_text = result.get("text", "")
@@ -229,13 +238,13 @@ def retriever(question: str):
     final_response = generate_response_with_prompt(combined_context, question)
     return final_response
 
-
-# Function to handle the entire audio query and response process
 def process_audio_query(audio_input, state):
     stream = None
-    # Process the audio input and get the transcription
     _, transcription, _ = transcribe_function(stream, audio_input)
-    print(f"Transcription: {transcription}")
+    
+    if not transcription.strip():
+        print("No valid transcription detected.")
+        return None, state  # Avoid generating a response for empty transcriptions
 
     # Retrieve a response based on the transcription
     response_text = retriever(transcription)
@@ -247,11 +256,9 @@ def process_audio_query(audio_input, state):
     # Update the conversation history in the state
     state.conversation.append((transcription, response_text))
     
-    # Return the path of the audio and the updated state
     return audio_path, state
 
 
-
 with gr.Blocks() as demo:
     with gr.Row():
         with gr.Column():