fromaudio_transcriberimportAudioTranscriberfromaudio_utilsimportdisplay_valid_input_devices,get_valid_input_devicesif__name__=="__main__":transcriber=AudioTranscriber()valid_devices=get_valid_input_devices()print("使用可能なオーディオデバイス:")display_valid_input_devices(valid_devices)# 対象のDeviceIndexを入力selected_device_index=int(input("対象のDeviceIndexを入力してください:"))# 文字起こしを開始transcriber.start_transcription(selected_device_index)

importasyncioimportqueuefromconcurrent.futuresimportThreadPoolExecutorimportnumpyasnpimportpyaudiofromaudio_utilsimportcreate_audio_streamfromvad_utilsimportVadWrapperfromwhisper_utilsimportWhisperModelWrapperclassAudioTranscriber:def__init__(self):self.model_wrapper=WhisperModelWrapper()self.vad_wrapper=VadWrapper()self.silent_chunks=0self.speech_buffer=[]self.audio_queue=queue.Queue()asyncdeftranscribe_audio(self):withThreadPoolExecutor()asexecutor:whileTrue:audio_data_np=awaitasyncio.get_event_loop().run_in_executor(executor,self.audio_queue.get)segments=awaitasyncio.get_event_loop().run_in_executor(executor,self.model_wrapper.transcribe,audio_data_np)forsegmentinsegments:print(segment.text)defprocess_audio(self,in_data,frame_count,time_info,status):is_speech=self.vad_wrapper.is_speech(in_data)ifis_speech:self.silent_chunks=0audio_data=np.frombuffer(in_data,dtype=np.int16)self.speech_buffer.append(audio_data)else:self.silent_chunks+=1if(notis_speechandself.silent_chunks>self.vad_wrapper.SILENT_CHUNKS_THRESHOLD):iflen(self.speech_buffer)>20:audio_data_np=np.concatenate(self.speech_buffer)self.speech_buffer.clear()self.audio_queue.put(audio_data_np)else:# noise clearself.speech_buffer.clear()return(in_data,pyaudio.paContinue)defstart_transcription(self,selected_device_index):stream=create_audio_stream(selected_device_index,self.process_audio)print("Listening...")asyncio.run(self.transcribe_audio())stream.start_stream()try:whileTrue:key=input("Enterキーを押したら終了します\n")ifnotkey:breakexceptKeyboardInterrupt:print("Interrupted.")finally:stream.stop_stream()stream.close()

importpyaudio# 有効なインプットデバイスリストを取得する関数defget_valid_input_devices():valid_devices=[]audio=pyaudio.PyAudio()device_count=audio.get_device_count()default_host_api_info=audio.get_default_host_api_info()default_host_api_index=default_host_api_info["index"]foriinrange(device_count):device_info=audio.get_device_info_by_index(i)if(device_info["maxInputChannels"]>0anddevice_info["hostApi"]==default_host_api_index):valid_devices.append(device_info)returnvalid_devices# 有効なインプットデバイスリストを表示する関数defdisplay_valid_input_devices(valid_devices):fordevice_infoinvalid_devices:print(f"DeviceIndex:{device_info['index']}, DeviceName:{device_info['name']}, 入力チャンネル数:{device_info['maxInputChannels']}")# オーディオストリームを作成する関数defcreate_audio_stream(selected_device_index,callback):RATE=16000CHUNK=480FORMAT=pyaudio.paInt16CHANNELS=1audio=pyaudio.PyAudio()stream=audio.open(format=FORMAT,channels=CHANNELS,rate=RATE,input=True,input_device_index=selected_device_index,frames_per_buffer=CHUNK,stream_callback=callback,)returnstream

importwebrtcvadclassVadWrapper:def__init__(self):self.vad=webrtcvad.Vad(3)self.RATE=16000self.SILENT_CHUNKS_THRESHOLD=20defis_speech(self,in_data):returnself.vad.is_speech(in_data,self.RATE)

fromfaster_whisperimportWhisperModelclassWhisperModelWrapper:def__init__(self):self.model_size_or_path="large-v2"self.model=WhisperModel(self.model_size_or_path,device="cuda",compute_type="float16")deftranscribe(self,audio):segments,_=self.model.transcribe(audio=audio,beam_size=3,language="ja",without_timestamps=True)returnsegments