Abstract: This paper studies briefly a distance measure for objective quality evaluation of speech communication channels using instantaneous and dynamic spectral features. It is based on studies for subjective and objective quality evaluation and LPC cepstral distance measure. The dynamic spectral feature is expressed by the cepstral regression coefficients. We used the three parameters: cepstral coefficients, cepstral regression coefficients and energy regression cofficients to measure objective speech quality. The experimental results show that the correlation coefficient between this objective measure and the subjective quality evaluation (MOS) is 0.953, it is much better than LPC cepstral distance measure. and other 2 measures.

Key words: objective quality evaluation, cepstral distance, regression coefficient

摘要： 主要讨论了一种使用语音频谱动态特性的倒谱距离测量对通信系统或通信设备进行客观音质评价方法。它是在使用LPC倒谱距离评价普通话语音质量的基础上发展起来的。语音频谱动态特性是利用倒谱对时间的回归分析来表示的。这种度量方法中，使用倒谱系数，倒谱回归系数和能量回归系数3个参数的距离测量；在使用了谱包络信息的基础上，考虑到了谱包络随时间的变化量及信号能量随时间的变化。比较了4种距离估算方法和主观音质评价(MOS)的相关性，实验结果表明：这种度量方法最好，明显优于LPC倒谱距离度量方法。

关键词: 客观音质评价, 倒谱距离, 语音频谱动态特性