基于唇动特征的语音端点检测

Speech endpoint detection based on lip moving

针对强噪声环境下语音端点检测问题,本文提出了一种利用唇动特征检测语音端点的新方法。首先进行人脸和嘴唇检测,然后分别用PCA法或DCT法对嘴部特征进行提取,再用概率神经网络进行唇动分类和识别。实验表明,在强噪声的环境下利用视觉通道的唇动特征进行语音端点检测是可行的。

To solve the problem of speech endpoint detection in strong noise environment , A new speech endpoint detection method based on the lip moving is proposed. First, find out face area and lip , then extract the lip feature using Principal component analysis or Discrete cosine transform method , and then classify and recognize lip based on Probabilitic neural networks. Experiment shows that speech endpoint detection based on lip moving is effective.