webrtcvad python——语音端点检测

py-webrtcvad 语音端点检测

算法说明

webrtc的vad使用GMM(Gaussian Mixture Mode)对语音和噪音建模，通过相应的概率来判断语音和噪声，这种算法的优点是它是无监督的，不需要严格的训练。GMM的噪声和语音模型如下：

p(xk|z,rk)={1/sqrt(2*pi*sita^2)} * exp{ - (xk-uz) ^2/(2 * sita ^2 )} 

xk是选取的特征量，在webrtc的VAD中具体是指子带能量，rk是包括均值uz和方差sita的参数集合。z=0，代表噪声，z=1代表语音

webrtc中的vad的c代码的详细步骤如下：

• 1.设定模式

  根据hangover、单独判决和全局判决门限将VAD检测模式分为以下4类

  0-quality mode
  1-low bitrate mode
  2-aggressive mode
  3-very aggressive mode
  

• 2.webrtc的VAD只支持帧长10ms，20ms和30ms，为此事先要加以判断，不符合条件的返回-1

• 3.webrtc的VAD核心计算只支持8KHz采样率，所以当输入信号采样率为32KHz或者16KHz时都要先采样到8KHz

• 4.在8KHz采样率上分为两个步骤

  • 4.1 计算子带能量

    子带分为80~250Hz，250~500Hz，500~1000Hz，1000~2000Hz，2000~3000Hz，3000~4000Hz
    
    需要分别计算上述子带的能量feature_vector
    

  • 4.2通过高斯混合模型分别计算语音和非语音的概率，使用假设检验的方法确定信号的类型

    首先通过高斯模型计算假设检验中的H0和H1(c代码是用h0_test和h1_test表示)，通过门限判决vadflag
    
    然后更新概率计算所需要的语音均值(speech_means)、噪声的均值(noise_means)、语音方差(speech_stds)和噪声方差(noise_stds)
    

实例代码

import collections
import contextlib
import sys
import wave

import webrtcvad


def read_wave(path):
    with contextlib.closing(wave.open(path, 'rb')) as wf:
        num_channels = wf.getnchannels()
        assert num_channels == 1
        sample_width = wf.getsampwidth()
        assert sample_width == 2
        sample_rate = wf.getframerate()
        assert sample_rate in (8000, 16000, 32000)
        pcm_data = wf.readframes(wf.getnframes())
        return pcm_data, sample_rate


def write_wave(path, audio, sample_rate):
    with contextlib.closing(wave.open(path, 'wb')) as wf:
        wf.setnchannels(1)
        wf.setsampwidth(2)
        wf.setframerate(sample_rate)
        wf.writeframes(audio)


class Frame(object):
    def __init__(self, bytes, timestamp, duration):
        self.bytes = bytes
        self.timestamp = timestamp
        self.duration = duration


def frame_generator(frame_duration_ms, audio, sample_rate):
    n = int(sample_rate * (frame_duration_ms / 1000.0) * 2)
    offset = 0
    timestamp = 0.0
    duration = (float(n) / sample_rate) / 2.0
    while offset + n < len(audio):
        yield Frame(audio[offset:offset + n], timestamp, duration)
        timestamp += duration
        offset += n


def vad_collector(sample_rate, frame_duration_ms,
                  padding_duration_ms, vad, frames):
    num_padding_frames = int(padding_duration_ms / frame_duration_ms)
    ring_buffer = collections.deque(maxlen=num_padding_frames)
    triggered = False
    voiced_frames = []
    for frame in frames:
        sys.stdout.write(
            '1' if vad.is_speech(frame.bytes, sample_rate) else '0')
        if not triggered:
            ring_buffer.append(frame)
            num_voiced = len([f for f in ring_buffer
                              if vad.is_speech(f.bytes, sample_rate)])
            if num_voiced > 0.9 * ring_buffer.maxlen:
                sys.stdout.write('+(%s)' % (ring_buffer[0].timestamp,))
                triggered = True
                voiced_frames.extend(ring_buffer)
                ring_buffer.clear()
        else:
            voiced_frames.append(frame)
            ring_buffer.append(frame)
            num_unvoiced = len([f for f in ring_buffer
                                if not vad.is_speech(f.bytes, sample_rate)])
            if num_unvoiced > 0.9 * ring_buffer.maxlen:
                sys.stdout.write('-(%s)' % (frame.timestamp + frame.duration))
                triggered = False
                yield b''.join([f.bytes for f in voiced_frames])
                ring_buffer.clear()
                voiced_frames = []
    if triggered:
        sys.stdout.write('-(%s)' % (frame.timestamp + frame.duration))
    sys.stdout.write('\n')
    if voiced_frames:
        yield b''.join([f.bytes for f in voiced_frames])


def main(args):
    if len(args) != 2:
        sys.stderr.write(
            'Usage: example.py <aggressiveness> <path to wav file>\n')
        sys.exit(1)
    audio, sample_rate = read_wave(args[1])
    vad = webrtcvad.Vad(int(args[0]))
    frames = frame_generator(30, audio, sample_rate)
    frames = list(frames)
    segments = vad_collector(sample_rate, 30, 300, vad, frames)
    for i, segment in enumerate(segments):
        path = 'chunk-%002d.wav' % (i,)
        print(' Writing %s' % (path,))
        write_wave(path, segment, sample_rate)


if __name__ == '__main__':
    main(sys.argv[1:])

参考：

http://blog.csdn.net/u012931018/article/details/16903027

GitHub地址：

https://github.com/wiseman/py-webrtcvad