[speech discrimination] discrimination of speech signals based on MATLAB double threshold method [including Matlab source code 1720]

One 、 How to get the code

How to get the code 1：
The complete code has been uploaded to my resources ：【 Speech discrimination 】 be based on matlab The double threshold method is used to distinguish speech signals 【 contain Matlab Source code 1720 period 】

How to get the code 2：
By subscribing to Ziji Shenguang blog Paid column , With proof of payment , Private Blogger , This code is available .

remarks ：
Subscribe to Ziji Shenguang blog Paid column , Free access to 1 Copy code （ The period of validity From the Subscription Date , Valid for three days ）;

Two 、 Brief introduction to short-term energy

1 Basic concepts
The double threshold method was originally proposed based on short-term average energy and short-term average zero crossing rate , The principle is that there are vowels in Chinese vowels , More energy , So we can find the finals from the short-term average energy , And the initials are consonants , They have a higher frequency , The corresponding short-time average zero crossing rate is large , So use these two features to find out the initials and finals , It is equal to finding out the complete Chinese syllable . The double threshold method is realized by two-level decision .

2 First class judgment
① According to a higher threshold selected on the speech short-term energy envelope ( threshold )T2( The figure is represented by a virtual horizontal line ) Make a rough judgment , Is higher than this T2 The threshold must be voice ( That is to say CD There must be voice between the paragraphs ), The starting and ending point of speech should be outside the time point corresponding to the intersection of the threshold and the short-term energy envelope ( That is to say CD Beyond the paragraph ).
② Determine a lower threshold on the average energy ( threshold )T( The figure is represented by a solid horizontal line ), And from Ｃ Point to left 、 from Ｄ Click right to search , Find the short-time energy envelope and threshold respectively T： Two intersecting points B and E, therefore BE Segment is the starting and ending position of the speech segment determined by the double threshold method according to the short-term energy .

3 Second judgment
Subject to the short-term average zero crossing rate , from B Point left and from E Click right to search , Find that the short-time average zero crossing rate is lower than a certain threshold ( threshold )T3 Two important points A and F( In the figure T3 Indicated by horizontal dotted line ), This is the beginning and end of the voice segment .
According to these two levels , The starting point of the speech is found A And end point position Ｆ. However, considering that there will be a minimum length in the mute area between words during pronunciation to indicate the pause between sounds , Is below the threshold T3 The end of the speech segment is judged only after such a minimum length is met , In fact, it is equivalent to extending the length of the final sound , Pictured 6-1-1 The starting and ending points of the voice indicated on the voice waveform diagram in are A and F+( It can be seen from the figure that the ending point is F, But in the actual treatment, it is extended to F+).
In the specific operation of endpoint detection , The first is to frame the speech ( The first 2 Chapter has been introduced ), The short-time average energy and short-time average zero crossing rate can be calculated on the basis of framing , Then compare and judge according to the threshold value frame by frame .

3、 ... and 、 Partial source code

clear all; clc; close all;
filedir=[];                             %  specify the path to a file 
filename='s.wav';                %  Specify a filename 
fle=[filedir filename]                  %  The string that makes up the path and file name 
[x,fs]=audioread(fle);                    %  Read in the data file 
x=x/max(abs(x));                        %  Amplitude normalization 
N=length(x);                            %  Take the signal length 
time=(0:N-1)/fs;                        %  computing time 
pos = get(gcf,'Position');              %  Make a picture 
set(gcf,'Position',[pos(1), pos(2)-100,pos(3),(pos(4)-200)]);
plot(time,x,'k');         
title(' Male voice “ The blue sky , White clouds , The green sea ” Endpoint detection ');
ylabel(' amplitude '); axis([0 max(time) -1 1]); grid;
xlabel(' Time /s');
wlen=200; inc=80;                       %  Framing parameters 
IS=0.1; overlap=wlen-inc;               %  Set up IS
NIS=fix((IS*fs-wlen)/inc +1);           %  Calculation NIS
fn=fix((N-wlen)/inc)+1;                 %  Find the number of frames 
frameTime=frame2time(fn, wlen, inc, fs);%  Calculate the time corresponding to each frame 
[voiceseg,vsl,SF,NF]=vad_ezm1(x,wlen,inc,NIS);  %  Endpoint detection 

Four 、 Running results

5、 ... and 、matlab Edition and references

1 matlab edition
2014a

2 reference
[1] Han Jiqing , Zhang Lei , Zheng tieran . Voice signal processing （ The first 3 edition ）[M]. tsinghua university press ,2019.
[2] Liu ruobian . Deep learning : Speech recognition technology practice [M]. tsinghua university press ,2019.
[3] Song Yunfei , Jiang zhancai , Wei Zhonghua . be based on MATLAB GUI Voice processing interface design [J]. Technology Information . 2013,(02)