0
$\begingroup$

So I know how to plot fft the audio file within Matlab and a window pops up. My question is, what next? I want to know how to go about measuring the frequency(s). I have two similar audio files, of two people saying the same thing, and i want to see if there's a relationship (correlation) between their voices in terms of frequencies and speed (bpm)

Improve this question
$\endgroup$
4
  • 1
    $\begingroup$ Welcome to DSP.SE! As the message above says, we need more information to try to figure out what you need to be answered. You have your signal x, you have its FFT X = fft(x). Why did you take the FFT? What information / processing are you trying to do? Your question doesn't make this clear. $\endgroup$
    – Peter K. ♦
    May 6, 2016 at 16:34
  • $\begingroup$ I want to know how to go about measuring the frequency(s). I have two similar audio files, of two people saying the same thing, and i want to see if there's a relationship (correlation) between their voices in terms of frequencies and speed (bpm). $\endgroup$
    – rez0nance
    May 6, 2016 at 16:56
  • $\begingroup$ Ok! Can you please edit your question to add this info? $\endgroup$
    – Peter K. ♦
    May 6, 2016 at 17:03
  • $\begingroup$ Done. Do you have any suggestions? $\endgroup$
    – rez0nance
    May 6, 2016 at 17:04

1 Answer 1

Reset to default
1
$\begingroup$

Speech processing is a very long and detailed topic, but let's make a start.

Correlation

This probably won't work, but the first thing you can try is cross-correlation of the signals.

xc = xcorr(speaker_1, speaker_2);

Note that really all you may get out of this is the delay between each sample (if you're lucky).

You might also get some level of correlation coefficient by taking the peak value of

xc = xcorr(speaker_1/std(speaker_1), speaker_2/std(speaker2));

Where Next?

If that doesn't work (I'm not hopeful), you will need to look at lots of other options:

  • Blocking the data into frames
  • Finding features within these frames (LPCs, etc.).
  • Using some method to time-align these between speech samples (DTW)
  • Finding some measure of "likeness" between the time-aligned features.
Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.