Abstract: To solve the?problem?of?information?loss under the traditional short-time analysis, a new wavelet packet transform algorithm based on the critical bandwidth is presented. The algorithm utilizes the process of MFCC parameters extraction with the introduction of the concept of critical bandwidth. Based on the GMM speaker recognition?system, the experimental results show the performance of the proposed algorithm is better than that of the traditional MFCC parameters in the condition of different wavelet?packet functions.

Key words: feature extraction, wavelet packet transform, critical bandwidth, Mel Frequency Cepstrum Coefficient（MFCC）

摘要： 针对传统语音短时分析技术容易出现丢失信息的现状，提出了一种基于临界带宽的小波包变换算法，该算法借鉴传统倒谱特征参数（MFCC）提取的过程并在该过程中引入临界带宽（Critical Bandwidth）的概念。在基于高斯混合模型的说话人识别系统中进行实验，结果表明在选取不同小波包函数的情况下，该算法所取得的识别率较MFCC参数均有提高。

关键词: 特征提取, 小波包变换, 临界频带, Mel频率倒谱系数（MFCC）