fft on samples of an audio file in matlab

I'm trying to extract information from a sound file in order to use it in a video classification algorithm I'm working on.

My problem is that I don't know how to work exactly with audio files in matlab.

Below is what I need to accomplish:

-open the audio file and get the sampling rate/frequency.

- I need to work on a window of 2 seconds so I have to loop over the file and get each 2 sec as a window and then do the ftt on each window.

After that it is my turn to use these values to do what I want

any help would be appreciated

Thanks.

Answer 1

[Carlos]

If you have a .wav audio named filename you can do

[y, Fs] = wavread(filename)

Where Fs is the sampling frecuency and y your audio samples.

Knowing the sample frecuency you can find the number of samples in 2 seconds (2*Fs).

Before performing the FFT algorithm of 2*Fs points you should apply some kind of windowing.You can find more about windows using doc window.

In order to find more about how to calculate the FFT type doc FFT in your Matlab command window.

[ir]

Actually, I'm working with mp3 file so I can't use wavread, I tried data = dsp. AudioFileReader("..."); but the length(data) is one, I'm not sure why it is like this. Also I used data = audioread(".."); but data does not seem a vector because I checked its size and it is [4186368 2], that means this audio file contains 2 channels do you have any idea about which channel should I use?

[Carlos]

It seems you are working with a stereo signal channel. You can sum the stereo signals and divide by two to obtain a mono signal.

Look at these other answers that deal with stereo to mono conversion.

http://www.mathworks.com/matlabcentral/newsreader/view_thread/44379

http://www.mathworks.com/matlabcentral/newsreader/view_thread/126472

[hemalatha]

please explain what is channel..and how the samples are separated in different channels

Answer 2

[ir]

I see, thanks for your help

Answer 3

[Theodoros Giannakopoulos]

Check out this code from the content of my book.

In general, Matlab does not provide mp3 IO support, so imo the best solution is to use a system command to call ffmpeg to convert the mp3s to wavs, before reading with wavread().

Regarding the window-fft-analysis task, there are many examples in the aforementioned code.