Based on Mel-frequency cepstral coefficients (MFCC),Mel-cepstral distance measure (Mel-CD) algorithm used for the objective evaluation of speech quality was analyzed.According to the theory of psychoacoustics,a human auditory model proposed by Johannesma and nonlinear compression were applied to extracting MFCC.Gammatone filter bank was used to simulate the basilar membrane.Mel-cepstral gammatone filter bank distance measure (Mel-GD) based on the improved MFCC was proposed,which was more in accordance with the auditory perceptual properties.Performance testing results showed that the proposed algorithm compared favorably with the Mel-CD in time complexity,the correlation degree between objective evaluation and subjective evaluation was improved by 4.9％,and estimation bias was decreased by 45.5％.%讨论了基于MFCC(Mel-frequency cepstral coefficients)特征参数的语音质量客观评价方法Mel-CD(Melcepstral distance measure).根据心理声学原理将Johannesma提出的人耳听觉模型和非线性压缩变换引入MFCC 特征参数的提取过程,用Gammatone滤波器组对人耳基底膜进行仿真.利用改进后的MFCC作为语音信号特征参数,提出了一种更加符合人耳听觉感知特性的客观评价方法——Mel-GD(Mel-cepstral gammatone filter bank distance measure).性能测试结果表明:所提算法与Mel-CD算法在时间复杂度上保持一致,评价结果的主观与客观的相关度提高了4.9％,平均估计偏差改善了45.5％.
展开▼
