I would like to know if it's possible to reconstruct the original time domain signal from it's time-stretched version? Is there any algorithm out there that can do this? Python, Matlab, etc?

I want to isolate certain sound events in the signal, however, the events are really close together, that it's hard to filter. A solution that I had in mind was to stretch the signal, filter this time-stretched signal and then reconstruct back by back stretching using inverse stretching factor to get the original length.

The algorithm will of course have to keep the correct pitch at all times. Is this possible or not?

  • $\begingroup$ It doesn't sound like a good approach at all. In fact, I think the idea is so far off that it's probably not even worth discussing. However, if you shared a better description of your original problem and you would almost certainly receive helpful suggestions for a practical solution. $\endgroup$ – Jazzmaniac Jul 8 '18 at 19:28
  • $\begingroup$ I am aware that is perhaps not a best approach, however my question is just to see if that is possible, not here to discuss about the best method. $\endgroup$ – Dan Jul 8 '18 at 20:43
  • 1
    $\begingroup$ Your question is too vague to give a better answer. If you insist on an answer, mine is "no". $\endgroup$ – Jazzmaniac Jul 8 '18 at 21:04
  • $\begingroup$ fake doppler. stretch or pitch but not both $\endgroup$ – user28715 Jul 9 '18 at 1:02
  • $\begingroup$ What sort of "sound events" are you trying to isolate? $\endgroup$ – A_A Jul 9 '18 at 7:00

Keeping your “correct pitch” after a time stretch can become a more difficult problem than your original problem of filtering short events. Both require a time-frequency resolution trade-off.


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.