复杂环境下基于准干净语音的音质评价方法

Quasi-clean speech-based speech quality evaluation method under complex environments

提出一种新的复杂环境噪声下无参考源语音质量客观评价方法.该方法基于准干净语音构造和有参考源评价模型,实现接近于有参考源客观评价的性能.首先,采用改进的最小值控制递归平均算法和多带谱减法获得带噪语音的准干净语音;然后,将该准干净语音作为改进的主观语音质量评估(PESQ)算法的参考语音,计算参考语音与带噪语音之间的失真误差,获得带噪语音的客观评价分值.实验结果表明:该算法客观评分相关度达到0.927,与有参考源语音质量客观评价标准的相关度0.931相比,实现了99%的接近,与无参考源的客观评价标准相比,提高了7.4%.

A new non-intrusive evaluation method for complex environments noise was proposed.The method was based on quasi-clean speech construction and intrusive perceptual assessment,and a similar performance to that of intrusive evaluation was achieved.Firstly,an improved minima controlled recursive averaging and a multi-band spectral subtraction algorithm were used to obtain the quasi-clean speech from the noisy speech.Then the quasi-clean speech was regarded as the reference speech to a modified version of perceptual evaluation of speech quality（PESQ）.The distortion between the quasi clean speech and the noisy speech was measured by the PESQ model,and the mean opinion score（MOS）was attained.Experimental results demonstrate that the proposed method gets a objective score correlation of 0.927,which is 99%similar to 0.931 of Reference source speech quality objective evaluation criteria,and 7.4% superior to objective evaluation criteria without reference source.