摘要
关于生物特征识别问题,人耳的听觉识别精度很重要。识别研究难点在于如何选取有效的耐噪特征参数,以提高识别率,传统的特征参数都将语音视为一种平稳信号进行处理,不能很好的反映语音信号的动态特性,故不能得到较好的识别率。针对提高抗噪声性能和识别声信精度,提出了一种新的特征参数(DWP-MFCC),用在感知倒谱分析(Mel-Cepstrum)的基础上引入多分辨率小波包分析技术,通过提高时频分辨率,增强语音动态信息,克服了原有单一线性分析的不足,并基于矢量量化(VQ)系统进行说话人识别实验。实验证明,与LPCC和MFCC参数相比采用新方法使系统的识别率得到显著的提高。
The challenge of voiceprint recognition is how to select an effective anti-noise feature parameters to improve the recognition rate.The traditional voice feature parameters which is regarded as a stationary signal processing do not well reflect the dynamic characteristics of speech signal,and it doesn’t acquire a good recognition rate.In view of this difficulty,the paper has proposed a new kind of speech feature（DWP-MFCC）.That is based on MelCepstrum Analysis and Multi-scale Wavelet Packet Analysis by raising time-frequency resolution and enhancing voice dynamic information,which overcomes the existing lack of a single linear analysis and carries out the speaker recognition experiments based on vector quantization（VQ） system.The experiment results show that this kind of feature has great immunity to the environmental noise.And the system built on DWP-MFCC has better identification accuracy than on LPCC and MFCC.
出处
《计算机仿真》
CSCD
北大核心
2010年第11期324-327,共4页
Computer Simulation
关键词
声纹识别
小波包分析
矢量量化
Voiceprint recognition
Wavelet packet analysis
Vector quantization
