Computer Engineering and Applications ›› 2011, Vol. 47 ›› Issue (10): 124-126.

• 数据库、信号与信息处理 • Previous Articles     Next Articles

Robust speech recognition based on static and dynamic feature parameters

ZHANG Junchang,LI Yanyan   

  1. School of Electronics and Information,Northwestern Poly-technical University,Xi’an 710072,China
  • Received:1900-01-01 Revised:1900-01-01 Online:2011-04-01 Published:2011-04-01

基于小波包分析的鲁棒性语音识别

张君昌,李艳艳   

  1. 西北工业大学 电子信息学院,西安 710072

Abstract: Through the analysis and research of Mel-Frequency Cepstral Coefficient(MFCC),can discover the limitation of using Fast Fourier Transform(FFT) which takes fix window width in the entire time and frequency space.It doesn’t match the characteristic of speech signal.But the wavelet transformation has the multi-resolution characteristic and can better conform to the auditory characteristic of human.This paper proposes a new speech recognition method DWPTMFCC of dynamic and static feature integration.Wavelet Packet Transformation(WPT) method is introduced to feature parameters in virtue of MFCC,then combines difference feature formed dynamic and static feature parameter integration.The simulation results show that the recognition rate is better using the new features than using MFCC in noise environment,especially in low SNR(Signal Noise Ratio).

Key words: speech recognition, Wavelet Packet Transform(WPT), feature extraction, Mel-Frequency Cepstral Coefficient(MFCC)

摘要: 通过对MFCC算法的研究,发现其中的FFT在整个时频空间使用固定的分析窗,这不符合语音信号的特性,而小波变换具有多分辨率特性,更符合人耳的听觉特性。提出了动静态特征参数结合的语音信号识别方法,首先在特征参数提取中引入了小波包变换,借助MFCC参数的提取方法,用小波包变换代替傅里叶变换和Mel滤波器组,提取了新的静态特征参数DWPTMFCC,然后把它与一阶DWPTMFCC差分参数相结合成一个向量,作为一帧语音信号的参数。仿真实验证明:基于新特征的识别率比原来MFCC的识别率有了很大提高,特别是在低信噪比情况下。

关键词: 语音识别, 小波包变换, 特征提取, Mel频率倒谱系数