用python实现语音端点检测(Voice Activity Detection,VAD)

1.准备环境
https://github.com/marsbroshok/VAD-python
里面的vad.py文件

2.具体代码

from vad import VoiceActivityDetector
import wave

if __name__ == "__main__":
    load_file = "test.wav"
    save_file = "process.wav"
    # 获取vad分割节点
    v = VoiceActivityDetector(load_file)
    raw_detection = v.detect_speech()
    speech_labels, point_labels = v.convert_windows_to_readible_labels(raw_detection)
    if len(point_labels) != 0:
        # 根据节点音频分割并连接
        data = v.data
        cut_data = []
        Fs = v.rate
        for start, end in point_labels:
            cut_data.extend(data[int(start):int(end)])

        
        # 保存音频
        f = wave.open(save_file, 'w')
        nframes = len(cut_data)
        f.setparams((1, 2, Fs, nframes, 'NONE', 'NONE'))  # 声道，字节数，采样频率，*，*
        wavdata = np.array(cut_data)
        wavdata = wavdata.astype(np.int16)
        f.writeframes(wavdata)  # outData
        f.close()

3.部分参数
vad.py文件

class VoiceActivityDetector():
    """ Use signal energy to detect voice activity in wav file """

    def __init__(self, wave_input_filename):
        self._read_wav(wave_input_filename)._convert_to_mono()
        #沿音频数据移动 20 毫秒的窗口。
        self.sample_window = 0.02  # 20 ms 
        self.sample_overlap = 0.01  # 10ms
        #应用长度为 0.5s 的中值滤波器来平滑检测到的语音区域。
        self.speech_window = 0.5  # half a second
        #计算语带能量与窗口总能量的比值。如果比率大于阈值（默认为 0.6），则将窗口标记为语音
        self.speech_energy_threshold = 0.6  # 60% of energy in voice band    
        #中值滤波器（滤波保留2000-8000hz）
        self.speech_start_band = 2000
        self.speech_end_band = 8000
        self.data_speech = []

4.实现效果
原音频

切割保留后的音频