0
$\begingroup$

The human voice is singular because it has a specific frequency spectrum and fast transitions between vowels and consonants.

What are good and computationally fast ways to detect if there is probably a voice contained in a noisy music, the amount of the song where there is vocals, and the type of vocals in different kinds of music, that perhaps have beats and guitars and other sounds mixed into the vocal?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
1
$\begingroup$

Scalable audio separation with light Kernel Additive Modelling

http://www.loria.fr/~aliutkus/kaml/

Improve this answer
$\endgroup$
10
  • $\begingroup$ Any chance for getting a descriptive answer? $\endgroup$
    – jojeck ♦
    Commented May 26, 2015 at 11:52
  • $\begingroup$ In addition it's hardly addressing the question. Vocal detection and source separation are only superficially related. $\endgroup$
    – Jazzmaniac
    Commented May 26, 2015 at 13:53
  • $\begingroup$ The linked tool separates vocals from beats and other music based on spectral patterns of those, it is exactly what author needs. You can also check examples here loria.fr/~aliutkus/kam $\endgroup$
    – Nikolay Shmyrev
    Commented May 26, 2015 at 15:09
  • $\begingroup$ @NikolayShmyrev, either you or me are misunderstanding the original question. In my eyes what he needs is not any form of separation, but merely presence detection of voice. Separating different sources provides little help there and overcomplicates the whole process. Even with separate sources, you'd still have to find some sort of classifier that tells you if one of these sources is voice. There is no reason to assume that such a classifier could not operate on the mixed source signal. In fact, a separation of different vocal features into different sources might make it harder. $\endgroup$
    – Jazzmaniac
    Commented May 26, 2015 at 18:11
  • $\begingroup$ Dear @Jazzmaniac, if you are interested in this technology you are welcome to visit the link and check the publication to learn how it finds voice source based on voice spectrotemporal properties. Method also detects beat sources and melody sources. The classifier is not required. $\endgroup$
    – Nikolay Shmyrev
    Commented May 26, 2015 at 19:26

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.