基于临界带特征矢量距离的端点检测算法

Voice Activity Detection Method Based on Selected Sub-bands Vector Distance

端点检测是语音数字信号处理中一个重要的环节。在前人研究的基础上提出了一种新的基于临界带特征矢量距离的端点检测方法,由计算得到的每帧各临界带中的功率谱之和作为特征矢量,并且通过计算各帧之间的矢量距离得到其距离轨迹,以此设定门限进行语音端点的检测。对比实验表明,相对于基于谱熵的算法及基于倒谱距离的算法,本方法具有更好的鲁棒性和较高的正确率。

Voice activity detected （VAD） is a very important step for speech signals processing, a new method of the VAD was proposed based on the selected sub-bands vector distance. It calculates the summation of power spectrum in very frame as characteristic vector. For detecting the voice activity, the method calculates the characteristic vector distances between frames to acquire the distance track. Theoretical analysis and experimentation show that compared with entropy-based algorithm and cepstrum-based algorithm this method has better robustness and higher correctness.