一种鲁棒的基于小波变换的语音参数提取算法 *

提出了一种基于小波变换的新型语音参数提取算法 ,提高语音识别系统对环境噪声的鲁棒性。由于引入了多分辨率小波分析技术 ,识别既在高频提供高的频率分辨又在低频提供高的时间分辨率。这样 ,提出的改进算法在语音词汇的识别更准确的同时 ,还大大简化了计算。将该算法和传统提取 MFCC系数的算法进行了比较,实验结果表明 ,利用小波计算语音特征具有更优的性能。

Robust algorithm for speech feature extraction based on wavelet transform

This paper proposed a novel approach for speech feature extraction, which promoted the robustness of speech recog-nition to noise. Wavelet transform was adopted to get both highest temporal resolution and frequency resolution in different po-sition. So the algorithm not only yield accurate measurements, but also exhibit a low computational cost. The experiment com-pared traditional MFCC algorithm and this method, results demonstrate that the proposed algorithm is robust and efficient for applications in speech recognition.