一种基于相对自相关序列的语音端点检测法

在研究单边自相关序列(One-sided autocorrelation,OSA)和相对自相关序列(Relative autocorrelation sequences,RAS)的基础上,提出了一种基于相对自相关序列的语音信号的端点检测算法。该方法利用相对自相关算法能够消除噪声的原理,以语音信号相对自相关序列短时平均幅度代替双门限比较法中的语音信号短时平均幅度,以语音信号短时平均幅度代替语音信号的短时平均过门限率,实验表明,在低信噪比下检测精度要高于传统的双门限比较法。

Speech endpoint detection based on relative autocorrelation sequences

In this paper,one-sided autocorrelation(OSA) and relative autocorrelation sequences(RAS) are studied,and a new algorithm of endpoint detection based on RAS is proposed.As relative autocor-relation has denoising capability,short-time average magnitude of speech signal in double-threshold com-parison is replaced with short-time average magnitude of its RAS,and the short-time average magnitude is substituted for short-time average threshold-crossing rate.Experiments show that the accuracy of endpoint detection is considerably higher than that obtained with the conventional method of double threshold comparison.