Update app.py

app.py

@@ -90,32 +90,23 @@ def auto_reset_state():
     time.sleep(2)
     return AppState()  # Reset the state
 
-# Function to process audio input and transcribe it
-def transcribe_function(state: AppState, new_chunk):
-    try:
-        sr, y = new_chunk[0], new_chunk[1]
-    except TypeError:
-        print(f"Error chunk structure: {type(new_chunk)}, content: {new_chunk}")
-        return state, ""
-
-    if y is None or len(y) == 0:
-        return state, ""
-
-    y = y.astype(np.float32)
-    max_abs_y = np.max(np.abs(y))
-    if max_abs_y > 0:
-        y = y / max_abs_y
-
-    if state.stream is not None and len(state.stream) > 0:
-        state.stream = np.concatenate([state.stream, y])
+# Function to process audio input and handle pause detection
+def process_audio(audio: tuple, state: AppState):
+    if state.stream is None:
+        state.stream = audio[1]
+        state.sampling_rate = audio[0]
     else:
-        state.stream = y
+        state.stream = np.concatenate((state.stream, audio[1]))
 
-    result = pipe_asr({"array": state.stream, "sampling_rate": sr}, return_timestamps=False)
-    full_text = result.get("text", "")
+    # Detect pauses in the audio stream
+    pause_detected = determine_pause(state.stream, state.sampling_rate, state)
+    state.pause_detected = pause_detected
+
+    # If a pause is detected and the user has started talking, stop recording
+    if state.pause_detected and state.started_talking:
+        return gr.Audio(recording=False), state
+    return None, state
 
-    threading.Thread(target=auto_reset_state).start()
-    return state, full_text
 
 
 # Function to generate a full-text search query for Neo4j
@@ -204,10 +195,14 @@ def retriever(question: str):
 
 # Function to handle the entire audio query and response process
 def process_audio_query(state: AppState, audio_input):
-    state, transcription = transcribe_function(state, audio_input)
-    response_text = retriever(transcription)
-    audio_path = generate_audio_elevenlabs(response_text)
-    return audio_path, state
+    state, _ = process_audio(audio_input, state)
+    if state.pause_detected:
+        # Perform transcription once pause is detected
+        transcription = pipe_asr({"array": state.stream, "sampling_rate": state.sampling_rate}, return_timestamps=False)["text"]
+        response_text = retriever(transcription)
+        audio_path = generate_audio_elevenlabs(response_text)
+        return audio_path, state
+    return None, state
 
 # Create Gradio interface for audio input and output
 with gr.Blocks() as interface:
@@ -219,4 +214,4 @@ with gr.Blocks() as interface:
     submit_button.click(fn=process_audio_query, inputs=[state, audio_input], outputs=[audio_output, state])
 
 # Launch the Gradio app
-interface.launch()
+interface.launch(show_error=True)