This paper proposes a modified Perceptual Evaluation of Speech Quality(PESQ) algorithm to improve the performance of objective speech quality evaluation. Syllable stability detection and Unvoiced/Voiced/Silence(UVS) classification are used in the proposed method. Parameters of stability of frames and syllables distortions are used to describe the effect to hearing perception, and these parameters are different to variant speech segments, especially to unvoiced, voiced and silence segments. Experimental results show that the proposed algorithm for PESQ is able to improve correlation results for narrowband speech and can help to improve other speech quality evaluation.%为提高语音质量客观评估的性能,提出一种改进的语音质量感知评估(PESQ)算法.该算法利用音节稳定性检测和清浊静音分类的方法,通过音节的帧间稳定性和损伤参数来描述语音听觉感知所受到的影响,这些参数对不同的语音段,如清音、浊音和静音具有不同的特性.实验结果表明,该算法能在窄带语音上提高PESQ得分与主观平均意见分的相关性.
展开▼
