一、简介
本文基于Matlab设计实现了一个文本相关的声纹识别系统，可以判定说话人身份。
1 系统原理
a.声纹识别
这两年随着人工智能的发展，不少手机App都推出了声纹锁的功能。这里面所采用的主要就是声纹识别相关的技术。声纹识别又叫说话人识别，它和语音识别存在一点差别。
b.梅尔频率倒谱系数（MFCC）
梅尔频率倒谱系数（Mel Frequency Cepstrum Coefficient, MFCC）是语音信号处理中最常用的语音信号特征之一。
实验观测发现人耳就像一个滤波器组一样，它只关注频谱上某些特定的频率。人耳的声音频率感知范围在频谱上的不遵循线性关系，而是在Mel频域上遵循近似线性关系。
梅尔频率倒谱系数考虑到了人类的听觉特征，先将线性频谱映射到基于听觉感知的Mel非线性频谱中，然后转换到倒谱上。普通频率转换到梅尔频率的关系式为：
c.矢量量化（VectorQuantization）
本系统利用矢量量化对提取的语音MFCC特征进行压缩。
VectorQuantization (VQ)是一种基于块编码规则的有损数据压缩方法。事实上，在 JPEG 和 MPEG-4 等多媒体压缩格式里都有 VQ 这一步。它的基本思想是：将若干个标量数据组构成一个矢量，然后在矢量空间给以整体量化，从而压缩了数据而不损失多少信息。
3 系统结构
本文整个系统的结构如下图：
–训练过程
首先对语音信号进行预处理，之后提取MFCC特征参数，利用矢量量化方法进行压缩，得到说话人发音的码本。同一说话人多次说同一内容，重复该训练过程，最终形成一个码本库。
–识别过程
在识别时，同样先对语音信号预处理，提取MFCC特征，比较本次特征和训练库码本之间的欧氏距离。当小于某个阈值，我们认定本次说话的说话人及说话内容与训练码本库中的一致，配对成功。
4 测试实验
可以看到只有说话人及说话内容与码本库完全一致时才会显示“密码正确”，否则显示“密码错误”，实现了声纹锁的相关功能。
二、源代码
function varargout = GUI(varargin)
gui_Singleton = 1;
gui_State = struct('gui_Name', mfilename, ...
'gui_Singleton', gui_Singleton, ...
'gui_OpeningFcn', @GUI_OpeningFcn, ...
'gui_OutputFcn', @GUI_OutputFcn, ...
'gui_LayoutFcn', [] , ...
'gui_Callback', []);
if nargin && ischar(varargin{1})
gui_State.gui_Callback = str2func(varargin{1});
end
if nargout
[varargout{1:nargout}] = gui_mainfcn(gui_State, varargin{:});
else
gui_mainfcn(gui_State, varargin{:});
end
% End initialization code - DO NOT EDIT
% --- Executes just before GUI is made visible.
function GUI_OpeningFcn(hObject, eventdata, handles, varargin)
% This function has no output args, see OutputFcn.
% varargin command line arguments to GUI (see VARARGIN)
% Choose default command line output for GUI
handles.output = hObject;
% Update handles structure
guidata(hObject, handles);
% UIWAIT makes GUI wait for user response (see UIRESUME)
% uiwait(handles.figure1);
% --- Outputs from this function are returned to the command line.
function varargout = GUI_OutputFcn(hObject, eventdata, handles)
% Get default command line output from handles structure
varargout{1} = handles.output;
% --- Executes on button press in trainrec.
function trainrec_Callback(hObject, eventdata, handles)
speaker_id = trainrec();
set(handles.train_current,'string','Hurraay,DONE!');
speaker_iden = sprintf('you re speaker number %d', speaker_id);
% set(handles.speaker,'string',speaker_iden);
set(handles.access,'BackgroundColor','blue');
set(handles.access,'string','YOU HAVE ACCESS, TRAIN COMMANDS NOW!');
% if access_ == 1
% set(handles.access,'string','YOU HAVE ACCESS, TRAIN COMMANDS NOW!');
% else
% set(handles.access,'string','YOU DONT HAVE ACCESS,SPEAKER NOT RECOGNIZED!');
% end
% --- Executes on button press in command.
function command_Callback(hObject, eventdata, handles)
trai_pairs=30;
out_neurons=5;
hid_neurons=6;
in_nodes=13;
eata=0.1;emax=0.001;q=1;e=0;lamda=.7; t=1;
load backp.mat W V;
recObj = audiorecorder;
Fs=8000;
Nseconds = 1;
while(1)
fprintf('say any word immediately after hitting enter');
input('');
recordblocking(recObj, 1);
x = getaudiodata(recObj);
[kk,g] = lpc(x,12);
Z=(kk);
Z=double(Z);
p1=max(Z);
Z=Z/p1;
for p=1:trai_pairs
z=transpose(Z(p,:));
% calculate output
y=(tansig(V*(z)));
o=(tansig(W*(y)));
break
end
b=o(1);
c=o(2);
d=o(3);
e=o(4);
f=o(5);
a= max(o);
if (b==a )
display('AHEAD');
set(handles.ahead,'BackgroundColor','green');
set(handles.command,'string','Ahead');
pause(2);
elseif (c== a)
display('STOP');
set(handles.stop,'BackgroundColor','green');
set(handles.command,'string','Stop');
pause(2);
elseif (d== a)
display('BACK');
set(handles.back,'BackgroundColor','green');
set(handles.command,'string','Back');
pause(2);
elseif (e==a)
display('LEFT');
set(handles.left,'BackgroundColor','green');
set(handles.command,'string','Left');
pause(2);
elseif (f==a)
display('RIGHT');
set(handles.right,'BackgroundColor','green');
set(handles.command,'string','Right');
pause(2);
end
set(handles.ahead,'BackgroundColor','white');
set(handles.left,'BackgroundColor','white');
set(handles.right,'BackgroundColor','white');
set(handles.stop,'BackgroundColor','white');
set(handles.back,'BackgroundColor','white');
end
function traincommands()
Fs=8000;
Nseconds = 1;
samp=6;
words=5;
recObj = audiorecorder;
aheaddir = 'C:\Users\Rezetane\Desktop\HRI Proj\Speech-Recognition-master\data\train_commands\ahead\';
backdir = 'C:\Users\Rezetane\Desktop\HRI Proj\Speech-Recognition-master\data\train_commands\back\';
stopdir = 'C:\Users\Rezetane\Desktop\HRI Proj\Speech-Recognition-master\data\train_commands\stop\';
rightdir = 'C:\Users\Rezetane\Desktop\HRI Proj\Speech-Recognition-master\data\train_commands\right\';
leftdir = 'C:\Users\Rezetane\Desktop\HRI Proj\Speech-Recognition-master\data\train_commands\left\';
s_right = numel(dir([rightdir '*.wav']));
for i= 1:1:samp
filename = sprintf('%ss%d.wav', aheaddir, i);
fprintf('Reading %ss%d ',aheaddir,i);
[x,Fs] = audioread(filename);
[s(i,:),g] = lpc(x,12);
end
for i= (samp+1):1:2*samp
filename = sprintf('%ss%d.wav', stopdir, i- samp);
fprintf('Reading %ss%d ',stopdir,i);
[x,Fs] = audioread(filename);
[s(i,:),g] = lpc(x,12);
%plot(s(i,:));
end
for i= (2*samp+1):1:3*samp
filename = sprintf('%ss%d.wav', backdir, i-2*samp);
fprintf('Reading %ss%d ',backdir,i);
[x,Fs] = audioread(filename);
[s(i,:),g] = lpc(x,12);
end
for i= (3*samp+1):1:4*samp
filename = sprintf('%ss%d.wav', leftdir, i-3*samp);
fprintf('Reading %ss%d ',leftdir,i);
[x,Fs] = audioread(filename);
[s(i,:),g] = lpc(x,12);
end
for i= (4*samp+1):1:5*samp
filename = sprintf('%ss%d.wav', rightdir, i- 4*samp);
fprintf('Reading %ss%d ',rightdir,i);
[x,Fs] = audioread(filename);
[s(i,:),g] = lpc(x,12);
end
S=zeros(1,13);
for i=1:1:samp
S=cat(1,S,s(i,:));
S=cat(1,S,s(samp+i,:));
S=cat(1,S,s(2*samp+i,:));
S=cat(1,S,s(3*samp+i,:));
S=cat(1,S,s(4*samp+i,:));
end
S(1,:)=[];
save speechp.mat S
trai_pairs=30; % 48 samples
out_neurons=5; % no of words
hid_neurons=6; %matka
in_nodes=13; %features are 13
eata=0.1;emax=0.001;q=1;e=0;lamda=.7; t=1;
load speechp.mat S
p1=max(max(S));
s=S/p1;
Z= double(s);
dummy=[1 -1 -1 -1 -1;
-1 1 -1 -1 -1;
-1 -1 1 -1 -1;
-1 -1 -1 1 -1;
-1 -1 -1 -1 1];
t=trai_pairs/out_neurons;
D=dummy;
for i= 1:1:5
D=cat(1,D,dummy);
end
三、运行结果
四、备注
版本：2014a
完整代码或代写加1564658423