Audionamix’s easy-to-use Mac utility promises to separate any music track into drums, music and vocals.
French DSP wizards Audionamix specialise in doing impossible things before petit dejeuner. Most of these relate to extracting individual voices and instruments from audio files that contain multiple sources, and this is the focus of their current flagship product. Reviewed in SOS December 2017, ADX Trax Pro 3 SP is designed to facilitate the cleanest possible extraction of a sung or spoken vocal from audio that also contains other sources, and to that end, offers a considerable degree of user control. The more help you can give Audionamix’s extraction algorithms, the more likely you’ll end up with a clean separation of vocal and backing.
However, not all of us can justify the outlay involved in buying Trax Pro 3 SP, or the time needed to fine-tune individual extractions; and Audionamix have now catered for those of us who just want near-instant gratification at an affordable price with the new Xtrax Stems. This is a Mac-only utility, though as with Audionamix’s other products, the actual processing is cloud-based, and you need to log in to your Audionamix account in order to use it.
In principle, Xtrax Stems couldn’t be easier to use. You simply drag and drop an audio file onto its single window, and select one of four processing algorithms labelled Automatic, Automatic HQ, Generic and Generic HQ. The file is uploaded to Audionamix’s servers, where their proprietary software gets to work on it, and three separate stem tracks are then returned to the application. Simple mixer controls then let you adjust the playback level and pan position of each stem, and there are also mute and solo controls.
While the upload and download times are usually pretty swift, the actual analysis can take several minutes if you choose one of the HQ algorithms. Sometimes there can be a wait before analysis starts, too, presumably while the queue at Audionamix’s end is dealt with. If you don’t like the way the first algorithm handled your audio, you can select a different one and have it analysed again, but you’ll be at the back of the queue.
The three stems you get back are always labelled Drums, Music and Vocals, which suggests that Xtrax Stems is mainly targeted at people who want to separate pop and rock tracks into their constituent tracks for remixing, or for educational purposes such as figuring out a tricky guitar part. Being the awkward sort, I naturally tried it out on a track that only contained acoustic guitar and vocals; understandably, it detected the guitar transients as Drums and the sustain portion of the notes as Music. Where the source material does contain the expected ingredients, Xtrax Stems usually does a reasonably good job of deciding what’s what, but it is not infallible, and quite often you’ll find that parts of a synth lead are treated as Vocals, or that backing vocals find themselves in the Music stem. If this happens, there’s not much you can do about it.
As to how good the results of the separation are, it really depends on what you’re hoping to achieve. The cardinal principle behind Audionamix’s technology is that as long as you don’t change anything, the individual stems will sum perfectly to recreate the original mix, and this is self-evident on playing back the track. But, of course, if you didn’t want to change anything, you wouldn’t be using Xtrax Stems; and the further you depart from the original mix, the more artifacts become apparent.
I thus found that Xtrax Stems was most successful in situations where you want to retrospectively rebalance a mix. For instance, there are plenty of ’60s recordings where the vocals are very high in the mix by modern standards. Xtrax Stems makes it easy to bring up the other sources by a few dB, with minimal side-effects, and in some cases the results definitely sounded better than the original.
At the other end of the spectrum, however, stems auditioned in isolation have obvious warbling artifacts, as well as occasional elements that have been mis-detected, so soloing the Drums stem certainly won’t give you something that sounds like a clean drum mix. I don’t think that any of my test material yielded individual stems that would be straightforwardly suitable for remixing, though with further work you could probably extract at least sections of a vocal that might be usable. The HQ algorithms generally sounded better than the non-HQ ones, and although Automatic usually sounded slightly different from Generic, there were not major differences in terms of detection and cleanness of separation. All in all, then, Xtrax Stems is dead simple to use, very affordable, and can be useful as long as you don’t expect too much of it!