I want to extract the similarities between 2 audio files. Here's an example so that you guys can get a better idea of what I mean:

I have 2 files, one where someone says "Duck goose" alongside some background music, and another file where the same person says "Duck goose" in the same exact manner, no differences at all, except the background music is different. How do I extract a "pure" version of the files where someone says "Duck Goose", with no background music at all. I want to do that by somehow comparing the two files, and extracting the similarities, those being a person saying "Duck Goose", in this case.

1 Answer 1


What you are proposing cannot be done. Your only option is to use something like the 'dialogue isolate' feature of Izotope RX to isolate the dialogue in each of the files, then compare one result with the other to pick the best outcome.

It is not possible to 'extract a pure version' of the dialogue because although the dialogue is 'similar', there is no actual correlation between the samples. Even if there were, this would only increase the dialogue level with respect to the background music and wouldn't remove it.

  • Actually, it can now be done. Just saw a video of some google researchers, literally 5 minutes ago, that have figured out a way to separate voices, but you do need to have video footage of the person saying 'duck goose' to make it work. ai.googleblog.com/2018/04/…
    – Mark
    Commented Jun 30, 2019 at 12:42

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.