Update app.py

app.py

@@ -132,31 +132,22 @@ def model(text, web_search):
         return "".join([response.token.text for response in stream if response.token.text != "</s>"])
 
 
-# Função assíncrona para obter as vozes
 async def get_voices():
-    voices = await list_voices()
+    voices = await edge_tts.list_voices()
     return list(voices)
 
-# Executar a função assíncrona
+# Executar a função assíncrona para obter as vozes
 voices = asyncio.run(get_voices())
 
 # Filtrar as vozes em português do Brasil
 pt_br_voices = [voice for voice in voices if voice["Locale"] == "pt-BR"]
 
-# Imprimir as vozes disponíveis em português do Brasil
-for voice in pt_br_voices:
-    print(f"Name: {voice['Name']}, Gender: {voice['Gender']}")
-
 # Escolher uma voz (por exemplo, a primeira da lista)
 chosen_voice = pt_br_voices[0]["Name"] if pt_br_voices else None
 
-if chosen_voice:
-    print(f"Voz escolhida: {chosen_voice}")
-else:
-    print("Nenhuma voz em português do Brasil encontrada.")
-
-    
 async def respond(audio, web_search):
+    if audio is None:
+        return None
     user = transcribe(audio)
     reply = model(user, web_search)
     if chosen_voice:
@@ -168,12 +159,31 @@ async def respond(audio, web_search):
         await communicate.save(tmp_path)
     return tmp_path
 
+def transcribe_and_respond(audio, web_search):
+    return asyncio.run(respond(audio, web_search))
+
 with gr.Blocks() as demo:    
     with gr.Row():
         web_search = gr.Checkbox(label="Web Search", value=False)
-        input = gr.Audio(label="User Input", sources="microphone", type="filepath")
-        output = gr.Audio(label="AI", autoplay=True)
-        gr.Interface(fn=respond, inputs=[input, web_search], outputs=[output], live=True)
+        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])
 
 if __name__ == "__main__":
     demo.queue(max_size=200).launch()