小词汇量孤立词语音识别系统多种特征组合参数的选择方法研究  

Research about the selection methods on various combination characteristic parameters in speech recognition system based on small vocabulary and isolated words

作　　者：张贺[1] 沈天飞[1] 滕秋霞[1] 

机构地区：[1]上海大学机电工程与自动化学院,上海200072

出　　处：《电子测量技术》2015年第3期48-53,共6页Electronic Measurement Technology

摘　　要：针对语音识别系统的多种语音特征参数进行了阐述,详细分析了语音特征参数中反映语音特性和说话人特征的美尔倒谱参数(MFCC)、差分美尔倒谱参数(ΔMFCC,Δ2 MFCC)、语音短时能量(Eg)以及语音基音周期参数(Ts)。在不同特征参数阶数下,将多种特征参数进行组合并将组合特征参数分别通过基于矢量量化(VQ)算法的语音识别系统进行验证,实验结果表明,在小词汇量孤立词语音识别系统中,基于MFCC、ΔMFCC、Eg和Ts的组合特征参数既能够较好的反应语音的动静态特性又对不同说话人的语音具有良好的区分度,为语音特征参数的研究提供了重要参考。The paper expounds the speech feature parameters of the speech recognition system,And it further analyses the Mel-scaled cepstrum coefficients （MFCC）.The differential MFCC （△MFCC,△2MFCC）,the speech short-time energy（Eg） and the speech pitch period coefficients （Ts）which reflect the speech features and speaker characteristics.In different characteristic parameters orders,a variety of characteristic parameters are combined and validated respectively in the speech recognition system based on Vector Quantization （VQ） algorithm.The experimental results show that,in small vocabulary and isolated words speech recognition system,the combined characteristic parameters base on MFCC、△MFCC、Eg and Ts can not only reflect the dynamic and static characteristics of the speech better but also has a good discrimination to different people＇s voice,which Provides an important reference to the research of speech feature parameters.

关 键 词：语音特征参数 MFCC 短时能量 基音周期 矢量量化 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]