一种基于感知特征动态失真度量的语音质量评估算法

摘要

This paper realizes an intrusive objective speech quality evaluation algorithm based on dynamic distortion measurement of speech perception characteristic parameters.The key steps of this algorithm(GFCCD_MOS) is feature extraction of Gammatone Frequency Cepstrum Coefficient,distortion measurement by Dynamic Time Warping and MOS-Mapping.This paper introduces the detail principle and measured the algorithm performance according to some indicators such as relevance and deviation error.The simulation results show the good performance of GFCCD_MOS.%实现一种基于语音感知特征参数动态规整失真度量的客观侵入式语音质量评估算法,该算法分为特征提取、失真度量、MOS映射三个步骤.算法的创新在于:特征提取过程中选取更能表征语音实质的GFCC参数取代传统的LPC、LPCC、MFCC、IMFCC等参数,在失真度量过程中选用动态规整距离取代传统的平均欧式距离,MOS映射时对映射函数进行修正以防止出现坏值而影响算法性能.文章详细介绍了算法的原理,在实现算法的基础上从相关度和偏离误差等指标对算法性能进行衡量,结果证明算法性能良好.