Convolution

Hello everybody

Please,I want to do the convolution on my speech signal using conv() How can I do that??

This is my signal

f=8000;

b=8;

[s1,f,b]=wavread('C:\Users\N\Desktop\sara.wav');

[Naz]

What do you want to convolve your signal with?

[sara s]

I know the convolution must be with something

But my prof ask me to do convolution only for speech signal and for that I'm confused

[Naz]

With a sinusoid of 8kHz? Something like modulation? Theoretically, if your voice contains frequencies over 4kHz you will have aliasing. Also, you need to choose how to truncate your sinusoidal signal (the length in time)

[Naz]

Convolution with itself? Then just do conv(s1,s1)

[sara s]

Yes,I did that previously

But I have this error

??? Error using ==> conv at 27

A and B must be vectors.

Answer 1

[Naz]

First of all, this

      f=8000;
      b=8;

is useless information because it will be rewritten anyway below, unless this information is given to you and you don't fully describe the required task. ASSUMING that your professor wants you to perform convolution of the recorded signal with itself, you can perform this:

[s1,f]=wavread('C:\Users\N\Desktop\family.wav');
s2 = conv(s1(1,:),s1(1,:)); %using all columns from the first row
subplot(2,1,1)
plot(s1(1,:))
subplot(2,1,2)
plot(s2)

Of course, the resulting signal will be nothing like the initial signal.

[sara s]

I have this error

??? Error using ==> conv at 27

A and B must be vectors.

[Naz]

Try replacing both s1 with s1'

[Walter Roberson]

Sara has a two-channel wav file so s1 is not a vector.

Answer 2

[Walter Roberson]

[conv(s1(:,1),s(:,1), 'same'), conv(s1(:,2), s1(:,2), 'same')]

That is for convolving each channel with itself. If for some reason you wanted to convolve left with right, then it would be

conv(s1(:,1), s1(:,2), 'same')

[Naz]

Good point

[sara s]

Thanks

But when I want to plot it,Nothing appear

[sara s]

I have this error

??? Error: An array for multiple LHS assignment cannot contain numeric value.

[Walter Roberson]

Please show the traceback of the error -- which line of code it occurred in, which routine, where it was called from, and so on.

[Timothy Dixon]

@ sara correct the type error from walter conv(s1(:,1),s(:,1)

correction

conv(s1(:,1),s1(:,1),

s1 was missing. if it does not work, then try to upgrade the version of Matlab that support the recording card system

Answer 3

[Image Analyst]

How about:

clc;    % Clear the command window.
close all;  % Close all figures (except those of imtool.)
imtool close all;  % Close all imtool figures.
clear;  % Erase all existing variables.
workspace;  % Make sure the workspace panel is showing.
fontSize = 24;

fullFileName = 'C:\Users\N\Desktop\sara.wav';

if exist(fullFileName, 'file')
  [s1,f,b]=wavread(fullFileName);
  subplot(2,1,1);
  plot(s1);
  grid on;
  title('Original Signal', 'FontSize', fontSize);
  windowSize = 201; % or whatever.
  s1_filtered = conv(s1, ones(1, windowSize ) / windowSize );
  subplot(2,1,2);
  plot(s1_filtered);
  grid on;
  title('Filtered Signal', 'FontSize', fontSize);
  % Enlarge figure to full screen.
  set(gcf, 'units','normalized','outerposition',[0 0 1 1]); % Maximize figure.
  set(gcf,'name','Demo by ImageAnalyst','numbertitle','off') 
else
  message = sprintf('File not found:\n%s', fullFileName);
  uiwait(warndlg(message));
end

[Walter Roberson]

I suspect Sara's wave file is two channel.

[Naz]

I suspect Sara gave up...

[sara s]

No, I'm not give up, but do you think it's true

Because when I asked my prof,he said that is not possible to make convolution between two signals and I'm confused from the information of my prof,all his information make me feel bother

[Naz]

Maybe his concern is in wording of convolution OF two signal with convolution BETWEEN two signals, but in the case of recorded voice, I would say it is very possible to perform convolution (with itself or any other real signal).

[sara s]

so I did that and do you think it's true

f=8000;

b=8;

[s1,f,b]=wavread('C:\Users\N\Desktop\family.wav');

s2 = zeros(size(s1));

for j1 = 1:size(s1,2)

s2(:,j1) = conv(s1(:,j1),s1(:,j1),'same');

end

plot(s2)

sound(s2,f)

[Walter Roberson]

That code looks like it should do convolution of the two channels against themselves. I have no idea what the mathematical result of that is expected to be, or what it would sound like.

Usually you would convolve against a much smaller vector to achieve a filter, such as convolving against [-1/2 1 -1/2] to get a variety of moving average.

There is probably a nice fourier analysis for what convolving a signal against itself would do, but I do not know what that analysis would be. Squaring all the frequency components, maybe ??

[Naz]

I just updated my answer

[Naz]

@Walter. Squaring of frequency spectrum

[sara s]

@Walter,Yes also I think that

[sara s]

Thanks so so much Walter and Naz

[Walter Roberson]

Wouldn't you need to square (the number of samples divided by 2) in order to square the frequency spectrum, if by that you mean that if you had 2*N+1 frequency bins in the fft before, that the new fft would have 2*(N^2)+1 bins and the new maximum frequency would be in bin N^2+1 ?

[Bjorn Gustavsson]

@Walter, convolving a signal with itself should give something close to the autocorrelation function - and that would be the Fourier transform of the power spectra.

[Image Analyst]

sara: Naz suggested that you'd want to convolve the signal with itself. I suggested just a box filter (local averaging) as an illustration. But YOU haven't yet said what is to be convolved with your signal. You also haven't said anything about needing frequencies, spectrum, Fourier analysis, or anything like that. Do you need anything like that or is any old signal convolved with your s1 good enough to complete your assignment?

[sara s]

Thanks a lot Image Analyst,that is enough for me :)

[sara s]

and I just wanted to make convolve the signal with it self and that is enough for me and Thanks a lot Image Analyst,you are very good

[Walter Roberson]

There are a number of different filters shown at http://web.mit.edu/1.130/WebDocs/1.130/Software/Examples/example1.m

In that code, you want the vectors that are directly underneath each comment. For example the line under '%Downsampling' is

x = [-1 0 9 16 9 0 -1] / 16;

and that x would be suitable to convolve against.

[sara s]

wow Walter thanks so so much ,that is so benefit

Thanks a lot