不同录音系统对声纹检测的影响  

Influence of recording system on voiceprint recognition

作　　者：达钊[1] 李倩[1] 郭霞生[1] 章东[1] 

机构地区：[1]近代声学教育部重点实验室南京大学声学研究所,南京210093

出　　处：《南京大学学报（自然科学版）》2011年第2期201-207,共7页Journal of Nanjing University（Natural Science）

基　　金：国家自然科学基金(10974093)

摘　　要：声纹识别已在身份识别中得以应用.本文采用三种常用数字录音系统(语音笔、话筒及手机)录制样本,并对这些样本进行基音分析,线性预测系数法提取共振峰以及美尔倒谱系数.结果表明,录音设备自身性能的差异对声纹参数存在影响,尤其是存在共振峰的丢失现象,提取的美尔倒谱系数包络存在一定差异.Voiceprint recognition has been applied in the identification field.With the widely use of digital recording systems,recorded voice samples processing and analysis become important,which leads to the study of the effect of recording systems on voice samples and the stable parameters.In this study,a variety of samples are recorded by three kinds of recording systems,i.e.voice pen,microphone and mobile phone.After speech preprocessing work,the sample pitch,linear prediction coefficients and the formant Mel cepstrum are extracted and analyzed.The results show that differences in recording systems affect the extracted parameters,especially the phenomenon of resonance peak loss,Mel Frequency Cepstrum Coefficient parameter differences.The pitch value and MFCC differences between the three recording systems are not that great as linear prediction coefficients.It is suggested that(1) high-quality recording equipment has optimistic formants lost phenomenon,and its bandwidth is obviously higher than low quality recording equipment.(2) Though MFCC parameters are robust,the MFCC information extracted from low quality recording equipment shows a weaker anti-noise ability.(3) Since different recording systems' influence on pitch is not very big,in dealing with low quality equipment recording speech samples,we can extract pitch parameters as well as MFCC parameters as the identification parameters.

关 键 词：声纹 线性预测系数 录音系统 美尔倒谱系数 共振峰 基音 

分 类 号：TN912.34[电子电信—通信与信息系统]