说话人识别中改进的MFCC参数提取方法

An Improved Method of MFCC Parameter Extraction in Speaker Recognition

在说话人识别技术中,特征参数的提取对语音训练和识别有着非常重要的作用。而Mel频标倒谱系数MFCC是一种常用的特征,它能对语音信号进行分析处理,去除对语音识别无关紧要的冗余信息,获得影响语音识别的重要信息。同时由于语音信号具有时变和混沌特性,以非线性随机共振理论和人类对听觉的理解为基础,提出了一种基于随机共振的MFCC特征参数提取方法。通过实验比较两种方法的结果,论证了改进方法的可行性以及优越性,为说话人识别技术中特征参数提取提供了一条新的研究方向。

Speech feature parameter extraction is an very important part of the speech recognition system,especially in speech training and recognition.Mel frequency cepstrum coefficient（MFCC） is a common feature,It can analysis and process speech signal,remove redundant information in speech recognition,and gain important information which influence speech recognition.Owing to time-varying and chaotic characteristic of voice signal,a improved MFCC feature extraction method based on nonlinear stochastic resonance theory is proposed.By comparison results of two methods,it is proved that the improved one is practicable and more superior which provides a new direction of speech feature parameter extraction in speech recognition.