Abstract: Mel-Frequency Cepstral Coefficients（MFCC） based on the human auditory system represents high recognition rate and strong power against noise compared with other features. However, due to the structure of its filter bank, it captures characteristics information more effectively in the lower frequency regions than the higher regions. Thus there must be some informations contained in the high frequency, which are missed. This work uses a new set of features by reversal of the filter bank structure which can make up the lack of MFCC. Considering the advantages of the two features MFCC and R-MFCC and using the Fisher criterion which is used to measure the recognition of various parameters, a new hybrid parameter is constructed through a combination of the Fisher criterion. Support vector machine as classifiers are adopted to identify speaker with MFCC, R-MFCC and the new hybrid parameter respectively. Experimental data shows that the new hybrid feature based on Fisher criterion is effective in raising the recognition rate of the speaker recognition.

Key words: speaker recognition, reversal Mel-Frequency Cepstral Coefficients（MFCC）, Fisher criterion, support vector machine

摘要： 基于人耳听觉感知的MFCC较其他说话人特征具有强抗噪性、高识别率特点。考虑美尔滤波器组的结构，其只在低频区具有较高的分辨率，在高频区分辨率却较低，这样势必会遗失一些包含在高频区域的重要信息。利用反美尔域下的特征R-MFCC与MFCC的各自优点，将R-MFCC与MFCC结合，形成优势互补，并给出了衡量各种特征参数识别能力的Fisher准则，结合Fisher准则构造出一种新的混合特征参数。采用支持向量机分别以MFCC、R-MFCC以及新构造的混合特征为参数进行说话人的识别，实验证明基于Fisher准则的优选混合特征作为说话人识别特征是可行的。

关键词: 说话人识别, 反美尔倒谱系数, Fisher准则, 支持向量机