论文部分内容阅读
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 studies the processing mechanism of the human ear’s acoustic signal at a certain level (mainly the cochlear and central auditory nervous system), and proposes an auditory model for speech recognition. The noise robustness of this auditory model is theoretically analyzed, and it is concluded that the speech recognition front-end processor built on this auditory model has good noise robustness. The same experimental results of linear prediction coefficient cepstrum (LPC-CEP) also prove that the noise robustness based on the speech feature of the auditory model is much better than the LPC-CEP feature.