extracting speech from audio

i want to extract 10 sections of a speech signal having spelled 1 to 10.in section i just get the 10th value i need all the values of all the spoken words. please help..

file = wavread( 'C:\Users\Desktop\samples\A');

%sound(file,11025);

totaltime = linspace(0,8,length(file));

i=0;

x=0;

y=0.8;

a=1;

b=8820;

section=(0);

while i<=9

time = linspace(x,y,8820);

%fourier = fft(file);

section1 = file(a:b,:);

section(i)=section1;

sound(section(i), 11025);

plot(time,section(i));

x=x+0.8;

y=y+0.8;

a = a + 8820;

b=b+8820;

i=i+1;

end

this is my project task.im new here. help needed.

Answer 1

[Walter Roberson]

That code should not run at all.

section1 is set to a slice of 8820 samples (in each channel). You then try to store that entire array into a single entry of a numeric array, section(i) . You cannot store 8820 (or 8820 by 2) numeric values into a single numeric location. Your code should exit.

Also, on the first iteration of the while, i is 0, and you are trying to store into section(i) which would be section(0) and that would crash because there is no element #0 in MATLAB arrays.

[faiza khan]

yes thats right.i dont know how to correct the looping.

[Walter Roberson]

I suggest you use

for i = 1 : 10

rather than the while loop.

Are the 10 sections each exactly 1/10th of the file? Or are they varying length and you need to figure out where the silence is? If you need to detect the silence, you will need to use a different structure for the code.

[Walter Roberson]

If the sections are of varying length then consider using cell arrays to hold them.

[faiza khan]

one speech file has 10 words spoken each word is spoken for 0.8 sec.lets say i have samples of 10 ppl.i need to make an array of all ppl's 1st word ,another array whose elements are the 2nd spoken word of all the ppl and so on..

[Walter Roberson]

For any one file, permute() and reshape() and permute() again, and store the result into a 4 dimensional array indexed by segment number, sample within segment, channel, and file number.

[faiza khan]

by any one file u mean the file having 10 spoken words from one person only.?

Answer 2

[Image Analyst]

That (if done right) only crops out chunks of the wavefile. It does not extract speech from an audio file -- like you asked for in your subject -- that has speech plus other unwanted sounds. I'm no audio expert but if you want to do that, you might try ICA to do blind source separation, as discussed in these web sites:

http://cnl.salk.edu/~tony/ica.html

http://research.ics.tkk.fi/ica/icademo/

http://research.ics.tkk.fi/ica/fastica/

Even if you did Fourier analysis, like you hinted at in your comment, this would simply do frequency filtering and wouldn't necessarily extract out speech from other sounds occupying that frequency range.

[faiza khan]

im loading 10 speech wav files into matlab..the first one loads correctly..in the other files the data duplicates.

[Image Analyst]

You mean the extracted output files, right - they are all the same? Try this:

Delete this line:

section=(0);

and change these lines from this:

section1 = file(a:b,:);

section(i)=section1;

sound(section(i), 11025);

plot(time,section(i));

to this:

thisSection = file(a:b,:);

sound(thisSection , 11025);

plot(time, thisSection);