基于听觉特性的声纹识别系统的研究  

Voiceprint recognition system based on auditory characteristics

作　　者：白燕燕[1] 胡晓霞[1] 郑三婷[1] 

机构地区：[1]西安工业大学北方信息工程学院,陕西西安710032

出　　处：《电子设计工程》2015年第4期85-87,91,共4页Electronic Design Engineering

基　　金：西安工业大学北方信息工程学院2013年度院长科研基金资助项目(YZ1308)

摘　　要：声纹识别技术,形象的说法就是说话人识别技术。它是根据人在说话时产生的波形,以及波形中反映人类心理和生理的特征参数来判断说话人的身份的技术。本文所研究的是与文本有关的说话人确认系统。比较了基于声道的线性预测倒谱系数(LPCC)和基于听觉特性的MEL频率倒谱系数(MFCC)参数特征,得出MFCC对环境存在更高的鲁棒性。并运用了隐形马尔可夫模型(HMM)在MATLAB上实现了语音数字的识别仿真。本实验系统的识别率达到了90%,验证了HMM模型识别的准确性。Voiceprint recognition technology is also called the speaker recognition technology with the figurative language.It is the technology to determine the identity of the speaker.It bases on the characteristic parameters of human psychological and physiological.These parameters are reflected in the waveform generated by the people when they speak.This paper is concerned with the text speaker verification system.It comes to a conclusion that the MFCC has a higher robustness to the environment to compare the Linear Prediction Cepstrum Coefficients（LPCC） based on the multichannel with MEL frequency cepstrum coefficients parameters characteristic（MFCC） based on auditory characteristics. And the use of Hidden Markov Model（HMM） simulation of digital speech recognition are implemented in MATLAB.This experiment system recognition rate achieved 90%,and verify the accuracy of the HMM model identification.

关 键 词：声纹识别 线性预测倒谱系数 MEL频率倒谱系数 隐形马尔可夫模型 

分 类 号：TN912.3[电子电信—通信与信息系统]