基于听觉模型的鲁棒性语音识别的研究

STUDY ON ROBUST SPEECH RECOGNITION BASED ON AUDITORY MODEL

本文在一定层次上(主要是耳蜗和中央听觉神经系统)对人耳的声信号的处理机理进行研究,提出了一种用于语音识别的听觉模型。从理论上对这种听觉模型的噪声鲁棒性进行了分析,得出建立在这种听觉模型之上的语音识别前端处理器具有良好的噪声鲁棒性。同线性预测系数倒谱(LPC-CEP)特征的对比实验结果也证明了基于听觉模型语音特征的噪声鲁棒性性要大大优于LPC-CEP特征。

This paper analyses the principle of speech signal processing of human auditory system (mainly of the cochlea and the central auditory system), proposes a new auditory model. Both theoretical and experimental study on noisy robustness of the representations of speech signal using this auditory model shows that this auditory model not only represents speech signal better, but also has good robustness.