基于Gammatone滤波器组的说话人识别算法研究  

Speaker recognition algorithm based on Gammatone filter bank

作　　者：茅正冲[1] 王正创 王丹[1] 

机构地区：[1]江南大学物联网工程学院,江苏无锡214122

出　　处：《计算机工程与应用》2015年第1期200-203,共4页Computer Engineering and Applications

基　　金：国家自然科学基金(No.60973095)

摘　　要：从人类听觉特性的角度出发,给出了一种基于人耳耳蜗听觉模型的Gammatone滤波器组模型。提出了一种新的Gammatone频率倒谱系数提取的算法,并在高斯混合模型识别系统中进行仿真实验。实验结果表明,采用这种新算法提取的Gammatone频率倒谱系数,应用于说话人识别系统,系统的识别率及鲁棒性都有明显的提高。From the perspective of human auditory characteristics, a Gammatone filter bank model based on the auditory system in human cochlea is presented. This paper proposes a new Gammatone frequency cepstrum coefficients extraction algorithm, and in the Gaussian mixture model recognition system for the simulation experiment. The experimental results show that using the new algorithm to extract Gammatone frequency cepstrum coefficients, applied in the speaker recognition system, the system recognition rate and robustness are significantly, improxed.

关 键 词：倒谱系数 高斯混合模型 识别率 鲁棒性 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]