一种基于语音分类的清晰度客观评价算法

Objective speech articulation evaluation algorithm using weighted Bark spectral distortion based on speech classification

为了能方便、快捷、可靠地对失真语音进行语音清晰度性能评价,提出了一种基于语音分类的加权巴克谱失真语音清晰度客观评价算法SC-W BSD。该算法提出在巴克域中对清音帧、浊音帧及过渡音帧巴克带谱失真进行加权来求失真语音的客观失真测度。通过分析3类语音的巴克谱失真测度与DRT分的相关程度,提出一组以相关系数的幂次方为权重的有效SC-W BSD权重矢量。实验结果表明,SC-W BSD方法和DRT分的相关度达到了0.924 73,与巴克谱失真测度算法相比提高了4%。

A new objective articulation evaluation method SC-WBSD （weighted Bark spectral distortion based on speech classification） was presented. The calculation of objective distortion measurement based on speech classification was proposed. Through the analysis of correlation between three speech classes Bark band distortion used in BSD and DRT, an effective Bark spectral distortion weight vector based on power of each class band correlation coefficients was presented. Experimental result shows that the correlation between SC-WBSD and DRT has arrived at 0. 92473. Compared with the BSD, SC-WBSD has a 4 percent increase in the correlation with DRT.