It has been a dream of musicians, producers, engineers and remixers for years, but now — according to technology company Audionamix — the ability to separate a fully mixed track into its constituent parts has finally been realised.
The complete algorithm for doing so, which Audionamix are calling ADX (Audio Dynamic eXtraction), is not publicly available, so for now the process is exclusive to their bespoke audio service. There are other examples of similarly pioneering techniques eventually becoming commonplace in the studio, however, such as Cedar Audio’s noise-reduction processors. These were originally exclusive to their bureau service but have since found their way into studios the world over, so it’s not unreasonable to assume that this new mix-separating technology will ultimately trickle down to the professional- and project-studio markets.
Audionamix call the software that they use to perform the task the Unmixing Station, and part of the reason for it not being sold to end users appears to be that the process isn’t entirely automatic. Before any of the spectral analysis involved has begun, the operator will split the track into very small parts, identifying any sections where instruments are playing solo, to provide a ‘fingerprint’ of the instruments for the software to ‘learn’ from. From there, the algorithms are tweaked and refined according to the programme material and the requests of the client.
Depending on the track, Audionamix claim to be able to achieve anything from vocal extraction and isolation to complete mix deconstruction, and they say the process generally takes between two and five days.
There is an intriguing demo of the software available on their web site, but Audionamix rightly point out that the results achieved using the demo software fall far short of those attainable through their tailor-made service. We’ve been playing with the demo at SOS towers and it would seem that the demo is a preset version of the algorithm, and doesn’t allow you to create your own ‘fingerprints’ for each individual instrument.
However, we did discover some very interesting things by phase-inverting some of the extracted audio and summing it with the original, and this certainly hinted at the software’s possibilities. In fact, one of the unique aspects of the Unmixing Station is that the sum of the separated audio is always identical to the original mix, and Audionamix say that this is key to the programme’s success. While we weren’t able to convincingly separate a complete mix ourselves using the preset demo, we have no doubt that a specialist engineer with the complete algorithm at his or her disposal could come up with some quite astonishing results.
Indeed, if you’ve seen the 2007 film La Vie En Rose, you’ll already have heard the software at work — Audionamix used their algorithm to deconstruct mono Edith Piaf recordings for the movie’s surround soundtrack. You can also hear an example of the Unmixing Station’s capabilities at http://audionamix.com/veralynn, where Vera Lynn’s vocals from her classic ‘We’ll Meet Again’ have been isolated and set against a brand new arrangement.
The cost of the service depends on a number of factors, including the source material, its format, and the extent of separation required by the client, but with the time and expertise involved in getting the best results, we suspect it won’t be cheap.
