基于盲源分离理论的麦克风阵列信号有音/无音检测方法

A Voice Activity Detection Method Based on Blind Source Separation for Microphone Array Signals

该文提出一种在方向性噪声场中多路麦克风信号同时进行有音/无音检测(VAD)的方法。在方向性噪声场中,由于各个麦克风接收信号中的噪声彼此之间相关,因而,可以利用盲源分离理论将方向噪声与语音源信号分离,从而获得相对比较纯净的语音源信号。对分离出的语音源信号进行有音/无音检测,获得VAD结果,同时估计出各个麦克风信号相对于该信号的时延值。以相对纯净语音源信号的VAD检测结果为参考,将其分别平移相应的时延值,即可同时获得多路麦克风信号的VAD结果。计算机模拟结果表明,在方向性噪声场的多种情况下,该方法对具有加性噪声的多路麦克风信号均具有较好的有音/无音检测能力。

A Voice Activity Detection （VAD） method for microphone array signals in directional noise field is proposed. As the noises received by different microphones are correlated with each other in directional noise field, relatively pure speech can be derived from any two array signals by using Blind Source Separation （BSS） method. The generalized correlation method is used to estimate time delay between this relatively pure signal and every channel signals of microphone array. In the same time, a long-term speech information method is applied to the relatively pure speech signal to obtain its VAD result. Then this VAD result is used as reference to produce those of all array signals by the time shifting of it according to each time delay values. Simulation results illustrate the validity of the proposed method.