摘要
在语音质量客观评估方法中,通常以Mel倒谱系数作为语音特征参数。但是Mel倒谱阶数的变化对评估效果有着直接的影响,同时Mel倒谱系数中各个分量对语音质量评估的贡献度不同,而以往的研究没有涉及Mel倒谱阶数和分量贡献度对语音质量评估效果的影响。实验确定了最佳阶数后,结合增减分量法对Mel倒谱系数进行贡献度的研究,通过对特征参数的选择重组,得到维数更低的IMFCC(Important Mel Frequency Cepstrum coefficient)。经检验,改进后的方法提高了2%的评估效果,降低了2%的预测误差。
In the objective evaluation method of speech quality,the Mel cepstrum coefficient is usually used as the speech characteristic parameter. However,the change of Mel's cepstrum order has direct influenceonthe evaluation effect,and the contribution of Mel cepstrum coefficients to speech quality evaluation is different.But the previous research does not relate to the effect of the above two factors on speech quality evaluation. The study of Mel cepstrum order on evaluation effect is made and the optimal order is determined. Then the contribution of Mel cepstrum coefficient is studied by combining the increment and decrease component method to get lower dimension IMFCC(Important Mel Frequency Cepstrum Coefficient). Test shows that the improved method promotes the evaluation effect by 2% and reduces the prediction error by 2%.
引文
[1]陈国,胡修林,张蕴玉,等.语音质量客观评价方法研究进展[J].电子学报,2001,29(4):548-552.
[2] BARNWELL T P. A Bush Statistical Correlation Be-tween Objective and Subjective Measures for SpeechQuality[C]//Proc 1978 IEEE ICASSP,USA:IEEE,1978:586-590.
[3] PICONE J W. Signal Modeling Techniques in SpeechRecognition[J]. Proceedings of the IEEE,1993,81(9):1215-1247.
[4]陈明义,孙冬梅,何孝月.基于改进MFCC语音特征参数的语音质量评估的研究[J].电路与系统学报,2009,14(3):111-116.
[5]赵凌伟,张磊.基于Mel尺度的语音干扰效果评估方法研究[J].无线电工程,2017,47(2):32-35.
[6]杨瑞田,周萍,杨青. TEO能量与Mel倒谱混合参数应用于说话人识别[J].计算机仿真,2017,34(8):215-219.
[7]胡航.现代语音信号处理[M].北京:电子工业出版社,2014.