Unmixing a track with AI assistance

Let’s use iZotope RX 7 Advanced’s AI-driven Music Rebalance module to break apart a musically dense sample lifted from a record…

01 >

We’re starting with a four-bar cycle of a track lifted from a vinyl record. This dense sample comprises a female vocal hook, drum groove, bass guitar and two guitar parts. We fire up iZotope’s RX 7 Advanced and drag this sample onto the main window to load it in.

02 >

The sample’s waveform can be seen on RX 7’s main spectral display. The software’s Music Rebalance module uses machine learning technology to analyse the audio and separate it into its constituent elements, which can then be turned up and down in the overall mix. Let’s load it up from the rightmost list.

03 >

The Music Rebalance module’s controls are relatively simple – all of the signal-splitting power goes on behind the scenes. The three Separation Algorithms determine whether the processing will be applied to the left and right channels together or separately. We decide to select the Advanced Joint Channel for best quality.

04 >

We pull down the Vocal slider to the minimum -Inf dB, then hit Render to print the results. Hey presto – when we play back the loop, the vocal has disappeared from the mix, leaving only a small amount of the vocal’s stereo delay remaining in the mix.

05 >

We hit Cmd-Z to undo this render, then head back to Music Rebalance. How about if we try and isolate the music and vocal in order to create a ‘remixable’, beatless loop? To do this, we pull the Drums slider down to minimum, then Render again. It worked!

06

Finally, let’s tune the results even further. We undo our previous render, then tweak the volume of the four elements until the vocal sits proudly on its own, with only a touch of guitar in the background. As you’d expect, there are some subtle ‘garbled’ artefacts, but they won’t be noticeable when we lay down a new beat and bassline over the loop.