I'd like to compare 2 speech similarity and show it in terms of 0-100%.
Here's my approach:

1: Convert 2 audio waveform into audio spectrum by Fast Fourier Transform (FFT) 2: Use Dynamic Time Warping to compare 2 spectrums

Everything is fine when converting them into audio spectrum But how to show the similarity in terms of percentage?

I've read this site: https://stackoverflow.com/questions/2168027/dynamic-time-warping-to-compare-two-audio-recordings

I'm working on exactly the same things with him. However I have no idea about the answer.

Could anyone please help me? Thanks!

  • 1
    $\begingroup$ If I want 0% to mean there's no similarity, and 100% to mean the 2 signals are identical. Is there any similarity measure that can output this percentage? $\endgroup$ – Amr Hesham Feb 13 '13 at 13:18

There are no doubt plenty of ways you could define one of the measures in the linked question, but the question remains, after doing so, is the measure in any way meaningful?

This all depends on what you intend to do with the metric after computing it: If it's a perceptual estimate (quality of a Codec, subjective similarity and so on), you will need a set of measured perceptual data to which you then attempt to fit (mathematically) your measure.


Your Answer

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