语音识别系统中语音活动性检测方法的研究

Research of Methods of Voice Activity Detection in Speech Recognition System

针对当前语音活动性检测技术中传统方法普适性差和在低信噪比下检测性能陡降的问题,研究了在低信噪比强噪声(平稳和非平稳)环境下的语音时频增强相和基于改进谱熵能量的活动性检测相结合的语音识别系统的研究。首先估计背景噪声能量,分别对语音信号进行频域和时域的增强处理;然后利用一种鲁棒性更好的特征参数来判断语音端点。验证结果表明,该方法在平稳和非平稳两类噪声环境下均具有较好的检测性能,其应用范围更广泛。

Aiming at the difficulties of the poor universality and the precipitation of detection performance at low SNR in the field of voice activity detection, this paper proposes a new voice activity detection method in high intensity noise （ stationary noise and non - stationary noise） environments based on modified entropy -energy and time -frequency enhancement. First of all, it estimates the environment noise and enhances the speech signal in frequency - domain and time - domaio. Secondly, a good robust feature parameter is employed to detect the speech endpoint. The experimentation result indicates that the algorithm has a better detection performance in stationary noise and non - stationary noise environments and a good universality.