【语音识别】基于MFCC的小波变换DTW实现说话人识别matlab代码
作者：互联网
1 简介
小波变换的发展为语音信号提供了新的处理方法与技术,从而使语音处理技术取得了较快的发展。说话人识别提取说话人的语音特征对说话人的身份进行确认或辨认。语音识别研究领域的一个重要研究方向,就是从语音信号中有效地提取个人特征信息进行说话人身份的识别。在说话人识别系统中,最重要的是能够从语音片断中提取代表说话人独有特征的稳定参数。基于短时傅立叶变换的常规特征参数构造表现出了很多不足,它是在短时平稳假定的基础上获取时频局部化信息,时频分辨率是固有的,其缺陷不能很好的处理语音信号这种非平稳信号。针对短时傅立叶分析的缺陷,本文通过对小波理论和说话人识别技术的研究,在学习传统的基于听觉机理的特征参数MFCC（Mel频域倒谱系数）的基础上,加上小波变换、小波包变换和小波多分辨分析,构造出了基于小波变换的说话人识别特征参数:WPTCC（小波包变换系数）。利用在Matlab平台上仿真软件构建的说话人识别系统,进行实验证明。理论和实验分析的结果可以说明利用小波变换提取的说话人识别特征参数取得了一定效果,相比于传统的分析方法确实具有更好的识别特性。语音识别率的提高需要对语音信号进行预处理,本文也利用小波变换的知识对待识别的说话人语音进行降噪处理和端点检测,以保证提取特征参数时的准确。针对小波分析的特点,小波分解预处理后,跳过信号重构,直接将分解分量进入以下一步处理,简化过程提高精度。
2 部分代码
close all;
chos=0;
possibility=5;
messaggio='Insert the number of set: each set determins a class. This set should include a number of speech for each person, with some variations in expression and in the lighting.';
while chos~=possibility,
chos=menu('speaker identification System','Select speech signal and add to database','Select speech signal for speaker identification','Delete database',...
'speech signal: visualization','Exit');
%--------------------------------------------------------------------------
%--------------------------------------------------------------------------
%--------------------------------------------------------------------------
% Calculate gmm of the speech and Add to Database
if chos==1
clc;
close all;
selezionato=0;
while selezionato==0
[namefile,pathname]=uigetfile({'*.wav','speech Files (*.wav)'},'Chose speech signal');
if namefile~=0
selezionato=1;
else
disp('Select a speech signal');
end
end
filt=melfilter(150,300,15);
fr1=frm(strcat(pathname,namefile),16,8000,1);
mc2=train(fr1,filt,20);
mc2=mc2(3:18,:);
mc1=banshengsin(mc2);
s1=pitch(pathname,namefile);
a=length(s1);
b=length(mc1(1,:));
if a>b
s1(b+1:a)=[];
else
s1(a+1:b)=0;
end
mc1=[mc1;s1];
[im is ip]=init(mc1,16);
[nim1 nis1 nip1 times]=gmm(im,is,ip,mc1);
data=struct('name',{},'means',{},'cov',{},'prob',{},'pitch',{});
if (exist('speech_database.dat')==2)
-----------------------
%--------------------------------------------------------------------------
%--------------------------------------------------------------------------
% speaker recognition
if chos==2
clc;
close all;
selezionato=0;
while selezionato==0
[namefile,pathname]=uigetfile({'*.wav','speech Files (*.wav)'},'Chose speech signal');
if namefile~=0
selezionato=1;
else
disp('Select a speech signal');
end
end
if (exist('speech_database.dat')==2)
load('speech_database.dat','-mat');
filt=melfilter(150,300,15);
fr=frm(strcat(pathname,namefile),16,8000,3);
l=length(fr(1,:));
nosp=length(data);
k=0;
b=0;
r=nosp;
while(r~=1)
r=floor(r/2);
k=k+1;
end
p(2,nosp)=0;p(1,1)=0;
for i=1:nosp
p(2,i)=i;
end
mc4=train(fr,filt,20);
mc4=mc4(3:18,:);
mc=banshengsin(mc4);
pitch2=pitch(pathname,namefile);
a=length(pitch2);
b=length(mc(1,:));
if a>b
pitch2(b+1:a)=[];
else
pitch2(a+1:b)=0;
end
mc=[mc;pitch2];
coff=length(mc(:,1));
o=length(mc(1,:));
frameparts=struct('frame',{});
s=mod(l,k);
y=floor(l/k);
if s==0
for i=1:k
frameparts(i).frame(coff,y)=0;
end
else
for i=1:s
frameparts(i).frame(coff,y+1)=0;
end
for i=s+1:k
frameparts(i).frame(coff,y)=0;
end
end
for r=1:k
count=1;
for i=r:k:l
frameparts(r).frame(:,count)=mc(:,i);
count=count+1;
end
end
c=length(data);
for i=1:k
% tic
p1=ident2(frameparts(i).frame,filt,data,p);
% toc
p=upd_pr(p,p1);
p=nmax1(p);
end
p2=p(1)/o;
scores=zeros(nosp,1);
for i=1:nosp
pitch1=data(i).pitch';
% tic
scores(i,1)=myDTW(pitch2,pitch1(1:length(pitch2)));
% toc
end
scores;
[m,n]=sort(scores);
b=p(2,1);
if or((p2>-25),b==n)
nm=data(b).name;
message=strcat('The speaker is : ',nm);
msgbox(message,'DataBase Info','help');
else
message='the speaker is a stranger.';
msgbox(message,'DataBase Info','help');
end
else
message='DataBase is empty. No check is possible.';
msgbox(message,'speech DataBase Error','warn');
end
end
%删除全部数据，或只删除一个人的数据
if chos==3
clc;
close all;
if (exist('speech_database.dat')==2)
load('speech_database.dat','-mat');
button = questdlg('which speaker do you want to delete?',...
'Genie Question',...
'all','specified','all');
if strcmp(button,'all')
delete('speech_database.dat');
msgbox('Database was succesfully removed from the current directory.','Database removed','help');
else
prompt={'Enter the name of speaker you want to delete'};
name='specified speaker delete';
numlines=1;
defaultanswer={'0'};
answer=inputdlg(prompt,name,numlines,defaultanswer);
nspeaker=length(data);
names=cell(1,nspeaker);
for i=1:nspeaker
names{1,i}=data(i).name;
end
[a,b]=ismember(answer{1,1},names);
if a==0
warndlg('the speaker is not exist.','Warining')
else
data(b)=[];
speaker_number=length(data);
save('speech_database.dat','data','speaker_number','-append');
message=strcat('you have succesfully removed The speaker : ',answer{1,1});
msgbox(message,'specified speaker removed','help');
end
end
else
warndlg('Database is empty.',' Warning ')
end
end
if chos==4
clc;
close all;
selezionato=0;
while selezionato==0
[namefile,pathname]=uigetfile({'*.wav','speech signal (*.wav)'},'Chose speech signal');
if namefile~=0
[x,fs]=wavread(strcat(pathname,namefile));
selezionato=1;
else
disp('Select a speech signal');
end
end
figure('Name','Selected speech signal');
plot(x);
pause;
x=trim(x);
plot(x);
pause;
filt=melfilter(150,300,15);
fr1=frm(strcat(pathname,namefile),16,8000,1);
size(fr1)
mc2=train(fr1,filt,20);
colormap(1-gray);
imagesc(mc2);
pause;
mc2=mc2(3:18,:);
imagesc(mc2);
pause;
y=bansin(16)
plot(y);
pause;
mc1=banshengsin(mc2);
imagesc(mc1);
pause;
% cor=CorrelogramArray(fr1,x,256);
% [pixels frames] = size(cor);
% colormap(1-gray);
%for j=1:frames
%imagesc(reshape(cor(:,j),pixels/256,256));
%drawnow;
%end
% pitch=CorrelogramPitch(cor,256,8000);
% plot(pitch)
s11=pitch(pathname,namefile);
plot(s11);
pause;
%[im is ip]=init(mc1,16);
% [nim1 nis1 nip1 times]=gmm(im,is,ip,mc1);
end
end
3 仿真结果
4 参考文献
[1]刘然. 基于小波变换的说话人识别应用研究. Diss. 西安电子科技大学, 2011.
标签：end,MFCC,signal,小波,length,speech,namefile,speaker,识别 来源： https://blog.csdn.net/qq_59747472/article/details/121433209