Another question our users often ask is “how to extract voice from songs ?”, “how can i make an acapella of this song” ? etc..
Vocal isolation is a black science, it’s the voodoo of audio processing… or just a good old spectral substraction !
The principle is basic, find a part of the song without the voice, and substract it from a part with the voice, so you are left with only the voice part. The ideal situation is where you have the normal song (voice + music) and an instrumental version of the same song.
The magic voice isolator plugin exists !! and it’s free.
Kn0ck0ut VST plugin is a free vocal isolation plugin, based on the spectral substracting principle. It takes two inputs : voice+instru on left and instru only on right and it performs a FFT processing for removing the instrumental part. A sort of acapella maker !
Here’s the VST plugin :
Now you can download the VST plugin and copy the dll in your VST folder.
How to isolate vocals / get the vocal out of the song :
If you have the normal song (vocal+instru) and the instumental version :
- 1. open Wavosaur
- 2. menu Files->Open and select the files to load (one song file + instrumental file)
- 3. if the files are stereo, for each file unselect left channel ( press the “1” key), select all (CTRL + A) and use menu -> Process -> Mute : this mutes the right channel
- 4. on the instrumental file, use menu->Process->Swap channels
- 5. use copy / cut / paste to select the part you want to “extract” and try to make the two files “in sync” the best you can (use zoom in in time & amplitude for more precision)
- 6. select all (CTRL + A) the instrumental part and copy (CTRL + C)
- 7. select all the normal part and use menu->Edit->Paste special-> Paste and mix
- 8. open VST rack, load KnockOut VST plugin, and open the plugin UI
- 9. use the “R input gain” to find the right balance between vocal isolation and overall quality : the more you increase the parameter, the more the vocal is isolated, but the effect is more drastic and you can clearly hear the artefacts of the spectral processing
if you don’t have an instrumental version, you can try to find a part in the song with no vocal, that is similar to the part with vocal.
here’s a test :
voice+instru on left & instru only on right:
after the kn0ck0ut processing for isolating vocal: