Analysis and processing of male and female voice signals based on MATLAB

List of articles

• ​ ​ One 、 Signal acquisition ​​
• ​ ​ Two 、 Voice signal processing ​​
• ​ ​ 3、 ... and 、 Finally, judge ​​
• ​ ​ Four 、 Advantages and disadvantages ​​

One 、 Signal acquisition

First of all, I use my own voice and the recording from a female classmate to save as boy.mp3 and girl.mp3, And then in matlab On the platform audioread The function remembers the sampling frequency and number of sampling points .

      
      

       
       file
       
       =
       
       'girl1.mp3'
       
       
[x,fs]
       
       =audioread(file);
      
      
      
      

       
       
1.
       
       
2.
      
      

among x Represents audio data , Audio data in the file , Return to m×n matrix , among m Is the number of audio samples read ,n Is the number of audio channels in the file .
fs Represents the sampling rate .

Two 、 Voice signal processing

Due to the different physiological structures of men and women , Men and women show different auditory characteristics , The pitch frequency of the male voice is about 100-200Hz Between . The pitch frequency of female voice is mostly 200—350Hz Between , In view of the obvious difference in pitch frequency between men and women , Therefore, basic audio can be used as the basis for male and female voice recognition .
therefore , In this paper, the self-made male and female voice signals collected above are extracted . From the beginning data=x(:,1) Extract mono , Use xcov（） Function to process the covariance of the data , The significance of covariance measures the degree to which each dimension deviates from its mean . If the value of covariance is positive , It shows that the two are positively correlated ( From the covariance can be derived “ The correlation coefficient ” The definition of ), A negative result indicates a negative correlation , If 0, That's what statistics say “ Are independent of each other ”. Calculate the frequency of the extracted audio through these three codes .

      
      

       
       yy
       
       =y(halflen: ylen);% Check to see the last segment of data 
       
       
[~,maxindex] 
       
       = max(yy);%yy The maximum of the matrix is passed out 
       
       

       
       fmax
       
       =fs/(maxindex
       
       +
       
       30);% Frequency calculation 
      
      
      
      

       
       
1.
       
       
2.
       
       
3.
      
      

3、 ... and 、 Finally, judge

Pass frequency 200 As a boundary to judge the voice of men and women ：

Four 、 Advantages and disadvantages

advantage ： This experiment simply judges the sound of men and women through frequency . More common use , The idea of the whole process is also very simple .

shortcoming ： Though we pass 200Hz As a judge, men and women , But personally, I don't think this is very reasonable . For example, some boys' voices are very much like girls' voices , So when we test, it is possible to recognize the voice of boys as girls , I have actually tested such data , So it's not very reasonable . If you want to get a more realistic judgment , We may need to use neural networks to train to recognize , The effect may be much better .