语音信号的Bark子波变换及其在语音识别中的应用

本文首先在语音感知实验基础上所得到的Bark尺度概念与子波的构造联系起来,提出了一种语音信号Bark子波及其变换.在感知特性中,它与人耳的听觉系统是直接吻合的:在数学上,它是一种非正交的但却是超完备的,可逆且是自反演的子波变换.应用在语音识别中时,利用子波的局部基性质,取较小的分析帧长,提高了前端处理的时间分辨率.实验表明,所形成的新特征在对辅音区别能力方面明显优于目前应用于广泛的MFCC特征.

Bark Wavelet Transform of Speech and Its Application in Speech Recognition

This paper firstly describes a novel perceptual wavelet and its transform of speech,which is based on Bark frequency scale concept coming from the speech perceptual experiment.In the nature of physics,it consists with human's cochlea filter properties;in the sense of mathematics,Bark wavelet transform is a kind of non orthonormal but overcomplete,reversible and self inverting wavelet transform.When it is applied in speech recognition,small analysis frame length is used by using the property of local base of wavelet transform,which can improve the time domain resolution of front end process.Experiment shows that the formed novel feature is superior to MFCC in the capability of distinguishing the consonants.