【摘要】 分析了国际电信联盟提出的语音质量感知评价算法的原理,优化了其中的心理声学模型,提出了高效语音质量感知评价(EPESQ)算法。该算法采用无滤波或理想带通滤波代替原算法的IRS滤波,用对数关系简化了强度扭曲过程,并基于人耳的掩蔽效应,按照语音帧能量的不同来简化扰动聚合的加权过程。实验表明,EPESQ算法的时间复杂度平均比原算法降低了39%,而与主观评价的相关度仅下降了8.8%。此算法适合于语音质量实时测评和嵌入式系统实现,并且已经应用在自行研制的Internet语音通信系统中。更多还原

【Abstract】 An objective evaluation algorithm of speech quality is proposed. It contributes 3 simplifications to the psychoacoustics model in ITU-T P.862 Recommendation: Intermediate Reference System (IRS) filter is removed or replaced by a rectangle filter. Intensity warping function based on Zwicker’s law is replaced by 2 kinds of linear functions of logarithmic power of speech frames. Weights over speech frames in disturbance integration procedure are reallocated in 1-0 pattern to simulate the mask effect of human hearing. The experiment results show that 39% of the floating points operations can be eliminated from P.862 with only a 8.8% decrease in average correlation to listener opinions. EPESQ algorithm has been implemented in our Internet voice communication system.更多还原