### 实现 Python 实时语音检测与静音切分 为了实现实时语音检测并根据静音片段分割音频，可以采用 `pyaudio` 库来捕获麦克风输入，并利用 `webrtcvad` 或者类似的库来进行静音检测。对于音频处理部分，则可借助于 `wave` 来保存切割后的音频文件。 #### 安装依赖包 首先安装必要的Python库： ```bash pip install pyaudio webrtcvad wave numpy ``` #### 音频捕捉与VAD配置 创建一个脚本来监听来自默认设备的音频流，并设置VAD参数以区分有声和无声区段[^1]。 ```python import pyaudio import webrtcvad import collections import contextlib import wave from pathlib import Path def record_audio(output_dir='output', vad_mode=3, chunk_duration_ms=30, silence_threshold_chunks=20, min_speech_chunk_length_ms=500): """Record audio and split it into segments based on VAD.""" output_path = Path(output_dir) output_path.mkdir(parents=True, exist_ok=True) FORMAT = pyaudio.paInt16 CHANNELS = 1 RATE = 16000 # Ensure this matches the expected rate by your model or processing needs. CHUNK_SIZE = int(RATE * (chunk_duration_ms / 1000.0)) vad = webrtcvad.Vad(vad_mode) # Higher mode means more aggressive detection. p = pyaudio.PyAudio() stream = p.open(format=FORMAT, channels=CHANNELS, rate=RATE, input=True, frames_per_buffer=CHUNK_SIZE) voiced_frames = [] ring_buffer = collections.deque(maxlen=int((silence_threshold_chunks*chunk_duration_ms)/min_speech_chunk_length_ms)) with contextlib.closing(wave.open(str(output_path/'temp.wav'), 'wb')) as wf: wf.setnchannels(CHANNELS) wf.setsampwidth(p.get_sample_size(FORMAT)) wf.setframerate(RATE) while True: try: frame = stream.read(CHUNK_SIZE) is_speech = vad.is_speech(frame, RATE) if not is_speech: ring_buffer.append(False) if all(not f for f in list(ring_buffer)): save_wave(voiced_frames, str(output_path/f'clip_{int(time.time())}.wav')) del voiced_frames[:] else: ring_buffer.clear() voiced_frames.extend([frame]) except KeyboardInterrupt: break stream.stop_stream() stream.close() p.terminate() def save_wave(frames, filename): """Save recorded data to a WAV file""" with contextlib.closing(wave.open(filename, 'wb')) as wf: nchannels = 1 sampwidth = 2 framerate = 16000 nframes = len(frames) wf.setparams((nchannels, sampwidth, framerate, nframes, "NONE", "not compressed")) wf.writeframes(b''.join(frames)) if __name__ == "__main__": import time record_audio() ``` 这段代码实现了从麦克风录制音频的功能，并通过 WebRTC 提供的 VAD 功能判断当前帧是否有声音存在。当连续多个帧被识别为无声音时（即达到了设定的沉默阈值），则认为一段完整的讲话已经结束，并将之前收集到的所有有声帧作为一个单独的剪辑保存下来[^2]。