I'm working on my thesis and it is about an speech enhancement system to use proper noise reduction method/algorithm for different types of noise.

I want to test and compare some filters/algorithms for different noise types added to speech and find the best filter/algorithm for each noise type(s) so the designed system would use the best algorithm when it detects the noise type.

Are there any methods to automatically detect noise type so a speech enhancement system switch the filter/algorithm? Considering noise type would change during the speech?

Any advise on noise classes and proper filter for each class?

  • $\begingroup$ what classes of noise have you identified you want to work with? This question, as is, is a bit too broad. $\endgroup$ Sep 2, 2019 at 22:07
  • $\begingroup$ @MarcusMüller Noises like babble, Helicopter,Bird,Communication channel,Car and... $\endgroup$
    – Filthy Man
    Sep 2, 2019 at 22:16
  • $\begingroup$ those are things from very different categories; can you structure these classes and edit your question to present a structured representation of your whole considerations so far, so that we don't feel like we're writing your thesis? $\endgroup$ Sep 2, 2019 at 23:23
  • $\begingroup$ @MarcusMüller That's not my main thesis contribution and you can skip answering my question if you don't feel good about answering it and try to speak for yourself next time, i simply asked a question and if didn't mention it was about my thesis you wouldn't know. $\endgroup$
    – Filthy Man
    Sep 3, 2019 at 3:41
  • $\begingroup$ Yes, but answering the question without you defining which classes of noise we're considering becomes content-wise as large as writing a thesis! That's what renders it too broad, imho. (I also don't think "adding a comprehensive list of the noise classes you want to consider" is too much to ask for if you're asking us to help with classification! Knowing what kind of things to classify really helps a lot...) $\endgroup$ Sep 3, 2019 at 7:31

1 Answer 1


Environmental Sound Classification might be suitable for what you are asking for. Datasets include Urbansound8k, ESC-50 and Audio set. Pretrained classifier like VGGish are available, which cover the 527 classes in Audioset.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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