with gr.Blocks() as demo:    
    with gr.Row():
        web_search = gr.Checkbox(label="Web Search", value=False)
        input_audio = gr.Audio(source="microphone", type="filepath", streaming=True)
        output_audio = gr.Audio(label="AI Response", autoplay=True)
    
    is_recording = gr.State(False)
    last_interaction_time = gr.State(time.time())

    def toggle_recording():
        return not is_recording.value

    def process_audio(audio, web_search, is_rec):
        current_time = time.time()
        if is_rec and (current_time - last_interaction_time.value > 2):
            last_interaction_time.value = current_time
            return transcribe_and_respond(audio, web_search), False
        return None, is_rec

    input_audio.stream(process_audio, inputs=[input_audio, web_search, is_recording], outputs=[output_audio, is_recording])

    demo.load(toggle_recording, outputs=[is_recording])

    # Carregando o arquivo JavaScript
demo.load(None, js="""
    async () => {
        try {
            const script = document.createElement('script');
            script.src = 'silence_detection.js';
            document.head.appendChild(script);
            console.log('Silence detection script loaded successfully');
        } catch (error) {
            console.error('Error loading silence detection script:', error);
        }
    }
""")

if __name__ == "__main__":
    demo.queue(max_size=200).launch()