Logic Pro: How To Use The Stem Splitter

Stem Splitter gives you the option to separate complete mixes into up to four parts: Vocals, Drums, Bass and Other.

We revisit some old tape masters with the help of Logic’s new Stem Splitter.

I covered the new intelligent Session Players back in the September issue, but the other headline feature of Logic Pro 11 is a surprisingly capable stem splitter. The less good news is that the stem splitting facility is only available to those running on an Apple Silicon machine, presumably because it makes use of the AI technology in the M‑series processors.

For those unfamiliar with stem splitting, the aim is to be able to extract individual parts from a complete mix, something that is accomplished on a spectral level and often achieved with the help of machine learning. Unsurprisingly, the results are never quite perfect, as even the best‑trained AI sometimes fails to separate the harmonics of one instrument from those of another, but the way the separation takes place means that if the individual stems are then added back together, the original audio file is recreated exactly — which means you can often usefully rebalance elements, even if the separation isn’t perfect.

Logic Pro’s Stem Splitter is to be found in the Functions menu and it can separate a track into a maximum of four parts, or ‘stems’, categorised as Vocals, Drums, Bass and Other: guitars, keyboards and so on fall into the Other category. If you have guitar and keyboards playing at the same time, you can’t extract the guitar track as you can with some of the more exotic (and more expensive) third‑party stem splitters.

Time To Split

To make this work, all you have to do is import a mix into Logic Pro 11, select it in the Main window and then go to Stem Splitter, which is at the top of the third section down in the Functions menu. If no audio track is selected, Stem Splitter will be greyed out. A window opens showing the four stem options with a tick box by each one, so if, for example, you just want to separate the vocals, you’d only tick the Vocals box. If you want the full four‑way split, tick all four boxes. Having done this, you click Split and in a surprisingly short period of time (typically half a minute for an average‑length song) your stems are displayed as four separate tracks within a Track Stack.

Once you’ve selected an audio track and invoked Stem Splitter, Logic will create a new Track Stack and populate it with the extracted stems.

In my own tests, examination of the individual tracks revealed that vocals are separated surprisingly well, though there are sometimes slight artefacts as the software puts some vocal information into the wrong stem. Drums also separate out very cleanly, with the elements most likely to suffer tonal changes being things like crash and ride cymbals — but it’s still pretty impressive, and these all come back cleanly when the stems are recombined. Likewise, you can mute the drum track and all traces of the drums vanish, leaving you free to add a new drum part if you want to. Bass guitar also splits out fairly cleanly, though on some occasions high notes or higher harmonics might end up in the Other section along with the regular guitars.

What Next?

If your plan is to extract vocals from a mix to use in your own compositions (assuming of course that you have the permission of the copyright holder), the results can vary depending on the character of the voice and on the type of mix you are extracting it from. If effects such as echo have been applied to the voice, those will also be extracted, often with a lesser fidelity than had the voice been reasonably clean.

On the flip side, if you mute the vocal track, you generally end up with a pretty clean backing track. However, for me that’s not the real power of stem splitting. As a test I found some recordings made by my own band dating from around 1978 to 1980, which had been copied onto cassette tape. All the original recordings and the machines that made them were long gone by now. These recordings were the result of my earlier adventures in home recording, the originals made on a Tascam four‑track open‑reel machine with the only effects being tape echo and a spring reverb. Back then, even mics such as the Shure SM58 seemed unaffordable, so I shudder to think what cheapies we pressed into service! Despite the ravages of age and the limitations of the gear available at the time, the recordings still sounded fairly clean, though the cheap mics lacked any decent bass response, so the tracks were definitely short on low‑end punch.

Combine & Conquer

As long as you plan to recombine all the tracks after processing, there’s a surprising amount that can be achieved without introducing noticeable side‑effects. For the vocals I used some gentle pitch correction and then a compressor (all courtesy of Logic Pro’s own plug‑ins), as we didn’t have anything so exotic as a compressor back then. I also added a little tape delay, as none had been used on the track I chose to experiment on.

For the drums I used compression plus some low EQ boost, and a similar combination was employed on the bass guitar, as that too was sounding distinctly thin. The Other section, which comprised my electric guitar and Roland GR‑300 guitar synth (sadly also long gone) was largely left alone, with just a little low‑mid EQ to help fatten it up. All this was done fairly quickly under the guise of experimentation, but when the four parts were played together, the track sounded warmer and more solid without any of my cheating trickery being obvious. I was also able to lift the vocal track by a dB or two.

I then did a few further experiments using Logic Pro’s Stereo Spreader plug‑in to add some width to the Other and Drums tracks, as the limitations of the time meant that pretty much everything was in mono. Though it has been around for quite a while, Logic Pro’s Stereo Spreader is actually quite a useful and often underappreciated tool, as you can change the position of the filter peaks by adjusting the frequency limits and the filter order, the latter determining how many peaks and troughs the filter will create. This allows the filters to be tuned by ear to a critical part of the sound being processed, for example the pitch of a kick drum. The process is mono‑compatible too, as whatever is done to one channel is applied in the inverse to the other, which is what creates the illusion of stereo width from a mono source. If the process is affecting the bass end too much, reducing the Lower Int fader to apply less processing at the low end of the spectrum is all that’s needed.

Summary

The ability to extract elements from a mix to reuse in new projects seems to be the headline use for stem splitting, but for me, the technology is most valuable when used to improving existing archive mixes, by allowing you to processes the stems of a mix individually before recombining them. As my example shows, it is also possible to add more of a sense of stereo width to mixes originally made in mono, both by adding stereo ambience/delay effects or by using artificial stereo widening techniques such as Stereo Spreader. Another valid application would be to, for example, remove the bass guitar parts and then record a new part to replace it, or take out the drums to use what’s left as the basis of a remix.

On my test mix, muting the bass guitar track left no noticeable trace of the bass guitar when the remaining three stems were recombined.

On my test mix, muting the bass guitar track left no noticeable trace of the bass guitar when the remaining three stems were recombined. It might also be possible to remove the Other stem so that you can build a new mix using the original vocal, drums and bass parts, though in my experience this is when vocal stem artefacts tend to show up the most, as small parts of the vocal spectrum end up being wrongly shunted into the Others track.

The success of this type of operation varies depending on the source material, so give it a try and let your ears decide. I still think that stem splitting is magic, though, and its side‑effects are far less than I would have believed possible.