Stem mixing: the future of digital DJ'ing?
I believe that watching a DJ perform live has got a lot less interesting since the move to mixing on laptops. Why is it that the DJ keeps asking us to put our hands in the air? Why is there more and more emphasis on big visuals instead of the music? Is it because both the DJ and crowd understand that it's really the computer doing the work, with automated beat and level matching? Watching someone on stage bent over a laptop, they could be checking their Facebook page, for all we know.
At the same time, many of the new DJ tools emulate the analogue past, even when the traditional constraints of mixing two stereo sources no longer apply. Just about all music is created as a multitrack recording, and yet digital DJs still work with stereo files, where most of the possibilities for creative mixing have been taken away. We can partially isolate a bass or vocal line with (somewhat over-used) EQ tricks, but we can't really separate the instruments in the stereo mix-down. This is why DJs have to hunt down 'clean' samples, such as 'a cappella' mixes, or four-bar drum breaks, and the same samples get used over and over again. And then again some more, until those samples become musical clichés.
Now audio is digital and downloaded, why can't every release be inherently remixable, whether that's live in the club or in another studio? We know that most digital music services are predicated on a business to consumer relationship in which audio files are lossy, sub-CD quality and ultimately disposable, but let's consider those specialist DJ labels and stores which sell music to people who actually care about music.
Audiophiles have already adopted the FLAC format, a royalty-free lossless codec which supports up to eight channels at higher than CD bit and sample rates. Support for multichannel FLAC in DJ tools would mean we could isolate four stereo stems (groups of mix channels) and therefore mix them independently. Every DJ release would then be an 'a cappella', a drum track, a bass line and a lead instrument track, but would remain playable in normal stereo.
Instead of buying a CD or vinyl release with some other remixer's ideas in three or four alternative versions, the DJ would now be free to remix as they choose, like Mad Professor remixing Marvin Gaye live on the radio. Then we might see some creativity and performance in DJ'ing again.
Of course, an eight channel FLAC file takes longer to download than a cruddy MP3, but it would be worth the wait because the DJ would quite literally be getting more (data) for their money. Broadband keeps getting faster, too. If you consider the market price of 12" vinyl singles versus CD singles or lossy downloads there's clearly a price premium for DJ formats. So there's a strong economic incentive for labels (especially the specialist dance labels) to support a multichannel DJ format.
If higher download times are a problem for some use cases, or the DJ just wants to preview the multichannel version before buying it, we have the option to provide a multichannel Ogg Vorbis file at CD bit and sample rate with lossy encoding. Guess what? The games industry already adopted multichannel Ogg Vorbis on the sly; this format is what makes multi-instrument games like the Rock Band series possible. These multichannel files are known colloquially as .mogg files, presumably standing for 'multitrack ogg', but in fact standard .ogg files already support up to 255 channels with the Vorbis codec. No doubt the fact that Ogg Vorbis is a royalty-free format made the adoption that much easier for the games industry.
Is it difficult to make multichannel FLAC and Ogg Vorbis files, or is proprietary software required? Not at all, you can make them very easily in Audacity. Here are the first demo files from me...
http://people.64studio.com/~daniel/DJ_Vadim-Saturday.ogg (lossy, 24.5MB)
http://people.64studio.com/~daniel/DJ_Vadim-Saturday.flac (lossless, 71.3MB)
These are made using the stem files from http://ccmixter.org/imaginashun dropped into Audacity, mixed down to four stereo tracks, then using the multichannel export option (in Edit -> Preferences -> Import/Export -> Use Custom Mix) to get eight channel output files. They are both exported as 44.1KHz sample rate, 16 bit depth, quality setting 5.
We need some way of labelling the stems so that we can create appropriate controls in our DJ software. The metadata for the demo .ogg file is as follows:
$ vorbiscomment -l DJ_Vadim-Saturday.ogg
ARTIST=Dj Vadim ft. Pugz Atomz
ALBUM=U Can't Lurn Imaginashun
COMMENTS=Four track stems
License=Creative Commons Attribution Non-Commercial 3.0
The main limitation of these formats is that we do not (yet) have any DJ tools that can play them back. If you use VLC to play the Ogg Vorbis file you can hear the effect of separate stems by switching from stereo to 5.1 outputs. (If you only hear the drums, it's because your media player is playing stem 1 and ignoring the others). I cannot get the multichannel FLAC file to play in VLC, mplayer or anything else so far without mixing it down to stereo. The Mixxx developers have drawn up a blueprint for what would be required to make this idea work.
One other thing we might need to address is a reference level for each stem. Since we have multichannel sources and we want to preserve the dynamics of the studio mix, we don't need to repeat the mistake of undue compression or hard limiting found with stereo files. Unlike in the days of vinyl, we can now automate the process of comparing mix levels and adopt a loudness standard such as EBU R128.
If you've got views on the direction that digital DJing should be taking, please post your comments here. We'll be taking these ideas to the Future Everything festival in Manchester, UK taking place on May 16-19 this year.