Our primary focus has always been to deliver the best-sounding music separations (STEMs) available anywhere. We are constantly developing and refining our algorithms and improving our separation platform. We have made 3 substantial upgrades to our products and deployed them free to our customers in the last 4 months alone. And there is lots more to come.
Since we launched the first versions of DeMIX Pro and DeMIX Essentials in 2018, there has been an explosion of companies offering demixing technologies. At present, the majority of these are using Spleeter under the hood to perform stem separation. This includes RX8, SpectralLayers, Acoustica 7, VirtualDJ, NeuralMix, and djayPro, as well as a multitude of websites offering per track separation. In contrast, at AudioSourceRE, we have always developed our own unique separation solutions for all our products, as well as our web-based site demixer.com.
While our separation solutions consistently score higher in technical metric tests used to evaluate separation performance, such as SiSEC, the only way to truly tell which is best is through listening tests. That’s why we were the first demixing company confident enough in our separation quality to offer free trials to potential customers. We conduct extensive user listening tests before every update.
Check out the comparison examples below and also what we look for in our STEM quality development.
|Original Mixture : Lucky Man by Red Revision –
You can clearly hear that our vocal algorithms successfully identify exactly when the vocal is actually present in the track, whereas Spleeter has lots of bleed from other sources when there is no vocal for example in the intro. Our vocals have also greater clarity and sparkle at the high end by comparison.
Drums & Bass
Our separated drums have much sharper transients and a tighter more natural sound, as well as less interference from other sources. On our bass separation, you can hear the attacks of the bass notes and much more of the articulation of the bass player.
Finally, the attacks of the instruments are much better preserved in our Other separation, giving a more natural sound.
A point to note is that in both DeMIX Pro and Essentials, you can change the order in which you do the separations. This can make a difference to the results, and a good starting point is to separate the most prominent source in the mix first. This will typically be vocals, but sometimes can be drums.
Apart from improvements in quality, we have also been making improvements in speed too. When we launched DeMIX Pro and Essentials back in 2018, our vocal separation algorithm took approx. 1.5 times the length of the song to separate. Since then, we have been making continuous improvements to our separation speeds, and now a vocal separation takes under 1/8th the length of the song to process. This means the separations are over 12 times faster, with further speed improvements due later in the year!
Remember you can get a free trial of DeMIX Pro if you want to run your own examples through our technology before you buy, and we also provide our powerful separation technology to developers and enterprise clients via our API.