I have no background in signal processing but I would like to know what is the actual state of the art for vocal elimination/reduction. Obviously, I am not talking about the infamous "stereo elimination", but other, more sophisticated techniques that not everybody is aware of.

I understand that for many projects there might be no software or online application available, but I'd find very useful to hear a demonstration.

  • 1
    $\begingroup$ i haven't heard "state of the art" since the 70's. $\endgroup$ – robert bristow-johnson Mar 9 '14 at 0:37
  • 2
    $\begingroup$ You may edit it in "da best stuff yooo" to make it more modern. :) $\endgroup$ – gd1 Mar 9 '14 at 9:39
  • $\begingroup$ well, it ain't from an actual product (since i have never worked on a vocal elimination product), but i imagine they look for a strong quasi-periodic signal with identical fundamental frequency in both left and right and use a tuned and tracking comb-filter on it to notch it out. they probably put in some restrictions on the fundamental frequency so that the bass is not eliminated. this will probably also eliminate a lead guitar solo, also. $\endgroup$ – robert bristow-johnson Mar 9 '14 at 13:45

Some recent ideas (2-3 years old, I haven't followed recently):

  • Design an elaborate generative model of the spectrogram of music signals, combining a non-negative mixture of a small set of background frames (models the background music) + a harmonic comb filtered by a non-negative mixture of smooth basis functions (the source filter model of the lead instrument), with constraints on the pitch transitions. Estimate model parameters from the audio clip. Use Wiener filtering to resynthesize the background music/vocals from the estimated spectra. See A musically motivated mid-level representation for pitch estimation and musical audio source separation by Durrieu et al. Such Bayesian approaches can be made more complex ad nauseam, by integrating prior knowledge about note transitions/durations - at the cost of increasingly hairy inference procedures. Demo time!
  • Techniques inspired by background subtraction in video analysis. The idea is that the accompaniment (especially in "bland" pop music) is invariant by a translation of a few bars (the duration of the "loop" or "riff") - while the vocals are not. Once the structure of the piece of music is known and can be described in terms of repeated groups of bars (a well-studied topic in itself), one can take the median of translated versions of the spectrogram to suppress the vocals. See REpeating Pattern Extraction Technique (REPET): A Simple Method for Music/Voice Separation by Rafil & Pardo. Demo time!

In the commercial space, you can download a trial for ADX Trax.


I looked at this about 5 years ago when I was first hosting karaoke. I didn't find anything that did anywhere near a decent job. The generally had a knob that would allow you to remove more or less vocal, but they all removed/garbled the instrumentals along with it.


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.