I’m including the full text of a specific method for extracting vocals from fully mixed tracks. This tutorial was posted to what is now the dead part of the gybo boards. This method is very hit or miss, but when it hits, it’s magic:
1. Get ur mp3 (make sure its atleast 192kps , preferably 256 or more) or WAV (much better obviously)
2. Load it into soundforge and do a ‘PAN/EXPAND’ process with the settings set to ‘mid side’ ..
This gives you a WAV with the Left-right material and centre mono material separated..
3. Grab the Centre mono material (which will hopefully have significantly more vocal than instruments/music.. save it as a mono WAV and load into Cool Edit
4. In Cool Edit select a section of instrumental with no vox (bit you want to reduce/remove from the vocal) and get a noise reduction profile (at 24000 FFT size + about 300 snapshots in the profile)…
5. then select the whole WAV.. preview the noise reduction.. listen to what it does..
6. In the eq graph line bit.. increase the noise red below 200 hz and less between 300-10000hz .. giving a bit more clarity to the main vocal ‘region’
7. Then maybe try some midband compression (300-8000hz) with a very fast attack (with readahead set to about 12ms) and 100ms release.. (i’ll have to post a pic of the comp curve to show you how to get good results) to help gate some of the background music left over.
8. Then go to www.soundhack.com and download the amazing, yet seemingly unheard of,.. spectral plugins (VST) (theres a free trial)
9. go into cubase (or wot ever prog u use) import the ‘pella’ you’ve made and use the spectral gate to reduce non fundamental harmonics (see soundhack pdf for more info)…
The trick is to do this ‘in the mix’ for best results… Using a fast attack and slow release with the spectral gate helps remove transients like snares and cymbals..
bloody hell .. can’t believe i’ve told you all how to do this.. took me bloody ages to perfect!..
Thanks for sharing Tim.