基于GFCC与RLS的说话人识别抗噪系统研究  

Speaker recognition anti-noise system research based on RLS and GFCC

作　　者：茅正冲 王正创 黄芳 

机构地区：[1]江南大学轻工过程先进控制教育部重点实验室,江苏无锡214122

出　　处：《计算机工程与应用》2015年第10期215-218,231,共5页Computer Engineering and Applications

基　　金：江苏省自然科学基金项目(No.BK20131107)

摘　　要：为了提高说话人识别抗噪系统的性能,提出了将RLS自适应滤波器作为语音信号去噪的预处理器,进一步提高语音信号的信噪比,再通过Gammatone滤波器组,对去噪后的说话人语音信号进行处理,提取说话人语音信号的特征参数GFCC,进而将特征参数GFCC用于说话人识别系统中。仿真实验在高斯混合模型识别系统中进行。实验结果表明,采用这种方法应用于说话人识别抗噪系统,系统的识别率及鲁棒性都有明显的提高。In order to improve the performance of speaker recognition system in noisy environment, a RLS adaptive filter as preprocessor to remove noise is presented, to further improve the SNR of speech signal, and then through the Gammatone filter bank to deal with the speech signal after denoising, extracting the feature parameter GFCC, which is used for speaker recognition. Simulation experiment is conducted with Gaussian mixture model based speaker recognition system. The experimental results show that the proposed algorithm significantly improves system recognition rate and robustness in noisy environment.

关 键 词：抗噪系统 递归式最小均方(RLS) Grammatone频率倒谱系数(GFCC) 识别率 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]