基于听觉模型的语音特征提取

Speech Feature Extraction Based on Auditory Model

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 speech processing is analyzed in three stages,the frequency decompose of cochlea,the energy exchange of inner hair cell and audit ory nerves and the parallel processing of Lateral Inhibition Network (LIN),and the auditory mathematics model is built. The feature for speech recognition is e xtracted from the auditory model. The spectrum of auditory model and LPC-CEP sh ows that this auditory model not only represents speech signal well,but also is noise robust. Some comparative experiment results of the feature based on audit ory model and LPC-CEP show that the auditory feature has better performance tha n LPC-CEP feature.