基于快速提升小波变换的语音特征提取方法

Speech feature extraction method based on fast lifting wavelet transform

为提高语音识别系统对环境噪声的鲁棒性,在快速提升小波的基础上,结合感知频域上的滤波与倒谱均值归一化技术,提出一种语音特征参数提取方法.仿真实验表明,与传统方法相比,噪声鲁棒性显著提高;在语音信号的信噪比相近情况下,与传统小波方法相比,该方法计算简便、易于编程、计算速度快.

To improve the noise robustness of speech recognition system, a speech feature extraction method is proposed based on fast lifting wavelet transform, which combines Mel-frequency filtering and cepstrum mean normalization and has good noise resistance. The simulation experiment shows that it can greatly improve the noise robustness than most other existing methods, and has the features such as convenient computation, simple implementation for programming, and fast computing speed compared with traditional wavelet methods.