论文部分内容阅读
分析了耳蜗对声音频率分解机理 ,毛细胞和听神经的能量转换机理以及中枢神经系统的侧抑制机理 ,分别在以上三个层次上建立了数学模型 ,并提取出识别语音特征参数。通过对听觉谱和L PC倒谱对比分析 ,得到了听觉谱适宜用作语音识别并具有良好的噪声鲁棒性的结论。听觉谱和L PC倒谱对比实验分析也反映了听觉模型特征的优良性能。
The mechanisms of cochlear frequency of sound decomposition, energy conversion of hair cells and auditory nerve, and lateral inhibition mechanism of central nervous system were analyzed. Mathematical models were established at the above three levels respectively and the parameters of speech recognition were extracted. Through the comparison of auditory spectrum and LPC cepstrum, the conclusion that auditory spectrum is suitable for speech recognition and has good noise robustness is obtained. Auditory and LPC cepstrum comparative experiments also reflect the superior performance of auditory model features.