The top waveform display represents the original mono mix of the musical theme from the movie Psycho. Below it are the separated stems after applying ADX.
For most projects, musicians ultimately think about mixing many audio tracks down to two or, at most, 5.1. But the other direction is also important sometimes: expanding an existing stereo or mono mix to multichannel. Of course, there are many algorithms that purport to do just that, but they rely on phase and level shifts to merely simulate surround sound.
What if you could run a stereo or mono audio file through your computer and extract individual tracks or instrumental sections? That's the idea behind ADX (Audio Dynamic eXtraction), a new DSP algorithm from Audionamix (audionamix.com). Developed during the course of five years, ADX might well propel us far beyond anything we know today in this regard.
The company is circumspect about exactly how ADX works. Suffice to say that it uses sophisticated DSP techniques to identify and isolate different types of sounds — such as vocals, bass, drums, sound effects and so on — within a mono or stereo mix. These tracks can then be processed and recombined as desired. Audionamix claims that all separated tracks are phase-consistent and add up exactly to the sum of their parts, preserving the integrity of the original recording.
ADX runs on standard workstations under Windows, Mac OS X or Linux. In its current state of development, the software takes days or even weeks to process a 1-hour file, depending on the complexity of the project — though, interestingly, stereo files take no longer than equivalent mono files.
The applications for such a process are many and varied. For example, older movies have a mono or stereo soundtrack, and when those films are restored and re-released on DVD and Blu-ray, consumers expect a 5.1 soundtrack. In 2007, Audionamix used ADX to separate the tracks in early 20th-century mono recordings of French chanteuse Edith Piaf for the surround soundtrack of the biopic La Vie en Rose. Other such projects have included François Truffaut's The 400 Blows and Wes Craven's A Nightmare on Elm Street.
The same process can be applied to older music recordings for which the original tracks are no longer available. Even if the original multitrack tapes can be found, they are often in poor condition, making it difficult if not impossible to remix from those original elements. Another important application is noise removal and restoration, which is accomplished by separating the intended sound from any unwanted elements, such as wind or crowd noise.
For now, Audionamix uses ADX to provide a service to copyright holders and remastering engineers, but the company also envisions a brisk business with documentarians and archivists. DJs and other musicians could engage Audionamix to extract various tracks from existing songs to remix or crossmix them into entirely new pieces. You send the original mono or stereo file to Audionamix, and the company returns a multitrack file that is compatible with professional audio systems such as Digidesign Pro Tools.
What about releasing ADX as a commercial product? This is a distinct possibility in the future, perhaps as a plug-in for Pro Tools and other common audio programs. In fact, you can download a stand-alone demo version from unmixingstation.com today. This limited and fully automatic version is optimized for acoustic instrumental music, such as jazz trios.
I've heard a few examples of what ADX does, and while they aren't perfect — I could still hear just a bit of other elements and some odd artifacts in the extracted tracks — these problems are easily masked in the remix, and the process allows far more precise manipulation of individual tracks within a mono or stereo mix than has ever been possible before. I hope to see ADX become available as a product in the music marketplace, where it will likely stir things up.