Extracting Vocals With iZotope Rx 8

iZotope Rx 8 has a mind-blowing feature called Music Rebalance that allows you to change the balance between Vocals, Bass, Percussion and Other instruments in an already-mixed track. Here’s a visual of my starting point, The Message by Grand Masterflash and The Furious Five:

I used Music Rebalance to mute the Bass, Percussion and Other instruments:

This is an incredible feature, since isolating the vocals from an existing track without using the original stems is an extremely complex feat of digital signal processing. That fact that it works at all is amazing, but it does introduce some pretty nasty artifacts; particularly in this track where we’ve got spoken-word vocals backed by a heavy beat with percussive and instrumental effects that sit all over the vocal frequency range.

The first thing I noticed is that the high frequencies are boosted, leaving the sibilants in the vocals sounding really harsh. I wanted to maintain some harshness consistent with the anger behind The Message, but not this much. I used RX 8’s De-ess function to tame it down a little:

The next obvious thing was that the vocals are drenched in reverb, and while this isn’t entirely a bad thing, it means I have no control over how much reverb I want it to have since I’m at the mercy of the original producer. Also, there are nasty reverb artifacts on the extracted vocals. RX 8 has another mind-blowing function called De-reverb, so I used it to remove most of the reverb. De-reverb also introduces artifacts of its own and a little bit of original reverb helps hide the remaining Music Rebalance artefacts so I had to compromise here:

Even after De-ess’ing, the remaining vocal had too much high frequency for my liking, so I used some EQ to get a more natural vocal sound like what I imagined the microphone feed in the studio sounded like. I kept it fairly subtle since I didn’t want to lose spectral content I can’t recover later and I’m going to be manipulating the sample heavily in Logic Pro X, which has its own EQ function. I want to focus here on functions that RX 8 can do which Logic Pro X can’t:

Finally, the remaining artifacts seemed to be happening mostly in the left or right channel but not both, so I used RX 8’s Center Extract to reduce widely panned artifacts. I was conservative with it though because I found too much left the vocal sounding lifeless:

The result sounds a little dull, but that’s primarily because it’s been stripped of its technomusical production effects. As a result, Mr Flash sounds like a regular angry dude ranting into a microphone rather than a hip hop gangsta. It also sounds like it’s been put through a low-frequency phaser, but I can’t see how to fix that. It’s not a bad effect though and it should be fine once I’ve smashed it with some granular synthesis in Logic Pro X.

Here’s what the extracted vocals look like in RX 8: