基于降维MFCC的语音年龄估测  

Speaker Age Estimation Based on Dimensionality Reduction MFCC

作　　者：刘益嘉 彭云祯 李滨彬 陈雪勤[1] LIU Yijia;PENG Yunzhen;LI Binbin;CHEN Xueqin(Soochow University, School of Electronic and Information Engineering, Suzhou 215006, China)

机构地区：[1]苏州大学电子信息学院,江苏苏州215006

出　　处：《电声技术》2018年第2期32-35,共4页Audio Engineering

基　　金：江苏省大学生创新创业项目省级指导项目基金支持

摘　　要：人类从幼年生长至老年的过程中,其语音特征会随年龄增长发生改变,因此通过分析说话人的语音信号可以估测说话人的年龄。关于语音年龄估测的研究目前较少,但计算量都普遍偏大。为了降低算法的复杂程度,方便其在Android平台以及其他物联网系统平台的移植,本课题提出一种采用降维的MFCC作为识别特征参数的语音年龄估测方式,将说话人按照年龄划段分组,从文本无关的语音中提取参数并建立高斯混合模型,通过模式识别来估测说话人所属的年龄段。实验测试表明,课题采用的降维方法基本不影响语音年龄估测的正确率,并且操作简单易行。Speaker age can be estimated because the phonetic features are always changing with people growing up. Although the study about speaker age estimation is not so much,the existing methods usually require an enormous calculated amount. This project intends to use one type of dimensionality reduction MFCC as the identification character to make it easier for the algorithm transplantation into the platform of networking equipment such as Android. We divide people into different age groups,calculate the dimensionality reduction MFCC from the text-independent speech and build the Gaussian Mixture Model for each group which is used to estimate the speaker age. The test that we did shows that,the identification character,the new dimensionality reduction MFCC,which is extremely easy to calculate,has a tiny influence on the accuracy rating of speaker age estimation.

关 键 词：说话人年龄估测 梅尔频率倒谱系数 高斯混合模型 降维 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]