Song Splitter App Achieves the Impossible in 40 Seconds

If you’re wondering why anyone would want to break up a perfectly good song, I have a simple answer.

I’m a bassist. If I want to learn Fleetwood Mac’s John McVie’s bass line DreamsI prefer to listen to that bass part alone. And once I have learned it, I want exactly the opposite: everything except the bass, so I could step into McVie’s shoes and play with the rest of the band.

Until very recently, neither was particularly easy. For most of the history of recorded music, the final mix really was final. Once the bass, drums, guitars and vocals were mixed in stereo, recovering the original ingredients was like trying to recover the RAW file from a JPEG.

Fortunately, that is no longer the case.

AudioPrism is a $79.99 (currently $59.99) Windows app that promises to separate finished recordings on your own PC, converting them back into individual tracks for vocals, bass, drums, and other instruments. You can also try converting those parts to MIDI and teaching yourself to play them through your KeyPrism trainer.

On paper, that’s almost exactly the practice tool you need.

And after trying it out, I discovered something quite unexpected: AudioPrism is very good at the parts that should be impossible and surprisingly bad at the parts that shouldn’t be.

From “impossible” to “done” in 40 seconds

The desire to separate recordings is certainly not new. In 2013, New Atlas reviewed Riffstation, software aimed at guitarists that could isolate parts of a finished recording and, crucially, allow players to solo or mute the instrument they were working on.

The separation technology was considerably more primitive, but the use case was already clear. Source separation has been in development for decades, but machine learning has transformed what is possible. Deezer’s Spleeter and Meta’s Demucs helped make neural network stem separation widely accessible, while Peter (the lord of the rings) Jackson’s team performed a spectacular overall demonstration of their MAL technology. Developed for The Beatles: Returnit was later used to separate instruments from the four-track masters for the 2022 remix of Revolverand finally untangling John Lennon’s voice from the piano on the old “Now And Then” demo.

Audio to MIDI followed a separate path. Ableton, for example, converted audio melodies, harmonies, and drums to MIDI with Live 9 more than a decade ago.

AudioPrism represents convergence: it demixes the recording, transcribes its constituent parts, and then learns to play them.

Better yet, and this is an element I personally find particularly attractive, the expensive computation is done on your own GPU. No uploading your music, no processing queues, and no cloud-based calculations. That’s not only good for privacy; It’s wonderfully immediate.

Amazing technology, puzzling product

I gave AudioPrism two very different tests. It was very easy to operate; just drag and drop an audio file into it.

Fleetwood Mac’s Dreams was the softball: an exceptionally clean studio recording with plenty of space around John McVie’s bass. AudioPrism separated the entire song into four tracks on my RTX 5070 Ti in less time than it would take you to spell “Fast Fourier Transform” out loud. Twelve-year-old me would have thought this was witchcraft.

The bass was superbly isolated, and when I put the four stems into Audacity, recombined them, and A/B compared the result to the original recording, the transition was invisible. I was really impressed.

Dreams It’s a fantastic track, but much less fun to play than with the genius of Tony Levin. So for the bastard test, I used red rain by Pedro Gabriel secret world live: A dense live recording filled with ambience, reverb and layered instruments.

The extracted bass had a lot more artifacts and an unpleasant resonant/hissing quality, but was definitely usable. More than enough for practice, a few minutes of cleaning would have left it spotless.

And that’s when AudioPrism started to drive me crazy. Having produced these excellent songs, where is the mixer?

As a bassist, my workflow couldn’t be more obvious. Solo the bass while I learn the part, then mute it and let me play with the rest of the band. Or similar for the chosen instrument.

Instead, I had to import them into Audacity and mute the bass myself. AudioPrism had performed the almost miraculous task of separating a finished recording and then had me abandon the application to do the trivial part.

As if to underscore the point, while finishing up this review I came across StemKit, another GPU-accelerated local stem splitter. And there, right in the app, was what I’d been asking AudioPrism for: a mixer. Separate level controls for each root, with integrated mute and solo buttons. There’s no need to resort to Audacity, and it’s free and open source!

It is not the only example. In January, New Atlas covered JBL’s $599 BandBox Trio, a program that uses AI stem separation specifically so musicians can isolate or mute instruments and play along with the rest of the track. In other words, the difficult technology serves the simple musical requirement, rather than becoming the product itself.

I didn’t have a good time with the MIDI side. Feeding AudioPrism with its own cleanliness Dreams The bass root produced unpredictable multi-part results rather than anything I found useful as a transcription. KeyPrism’s descending note playback also stuttered and glitched repeatedly, making meaningful assessment of its timing and accuracy difficult.

Then there is the interface. My God.

AudioPrism looks like it was designed by some awesome kid who just discovered Winamp viewers. Glowing particles, mouse trails, pulsating graphics, and a constantly moving hyperspace background compete for your attention; I couldn’t find a way to turn them off. And, as a final insult, the hyperspace animation isn’t even a clean loop!

I’m trying to concentrate. I’m trying to learn something. And he’s doing everything he can to deny me.

The impossible is no longer the problem

AudioPrism also faces formidable competition: the aforementioned Stemkit. Ultimate Vocal Remover is also free and open source, and supports technologies such as Demucs and MDX-Net. Moisés approaches the problem much more like a musician: separate tracks are placed in a mixer where they can simply be muted, soloed or adjusted, along with other practice tools. The problem is that it relies on cloud processing and the paid tiers expand what you can do. So, No local.

Even the advantage of local processing is no longer unique to AudioPrism. LALAL.AI’s Lyra model performs stem separation on-device via its desktop app and VST plugin, although local processing requires its $19.99/month Pro subscription rather than a one-time purchase.

And maybe that’s the real story here. We’ve reached the point where getting convincing bass out of a 50-year-old stereo recording in 40 seconds on a home PC is no longer the problem. What is built around that capacity is.

That’s what makes AudioPrism so frustrating. Their core separation technology can be spectacular, and doing all that work locally is exactly the kind of thing that floats my boat. But it repeatedly seems designed around what technology can do rather than what a musician really wants to do with it.

Give me AudioPrism’s unique cost, excellent local separation, a simple mixer with mute and solo buttons, reliable transcription, and an interface that doesn’t look like an ’80s arcade, and we’d be getting remarkably close to an ideal practice tool.

AudioPrism has already solved what seems like the impossible part. It’s the ordinary parts that aren’t there yet.

I have to go. Sanctified’s Pino Palladino line has just finished cooking…