You know Shazam and Sound hound right? These companies have tech that ‘listens’ to the music you send to them via your mobile and identifies the track for you so you can buy it. I have no idea on ‘how’ they do this precisely but my guess is that it’s a time/frequency content ‘mapping’ process where the sonic EQ signature at various intervals is measured against the ‘master’ copy on the database. Variance will be accounted for in the algorithm in order for it to compensate for noise in the venue and often they’ll get it right. I wonder whether also they are mapping the texture at the beginning and end of each track; are there beats before the music hits? is that just a pad and piano on the intro behind the vocalist? Are the beats filtered on the intro? If so we could be looking at the early form of a database that could change the shape of music production forever.
‘Big Data’ is slowly starting to hit the public’s consciousness. At the moment, companies such as Facebook and Google are at the epicentre of a staggeringly large amount of data gathering. Everything you do on the web is being logged and the database grows. In isolation much of this data will be pointless and irrelevant. When you think about this being multiplied by the amount of people using the web however, the amount of data is enormous and can be used by algorithms to spot trends, recommend better ads for you etc.. Most of you will have a basic understanding of how peer to peer file sharing works; multiple computers sharing the load of distributing a single file. If you’re a Apple Logic Pro user you may also have come across Logic Nodes. These allow you to spread the computation of various effects processes or instruments across multiple macs on the local network. The technology behind big data is similar where multiple computers are being used to process a huge amount of information quickly and efficiently.
Coming back to my questions about Shazam/Soundhounds sniffing algorithm, I wonder just how much data they are gathering and exactly what could be done with it? What if that database could be used to create music, algorithmically composed in a way that would mimic those who have released tracks that the public have loved. Firstly, the popularity of individual tracks IS being measured, and not just on Soundhound/Shazam. It’s also being monitored on every single music download and streaming site that exists. It’s not just used for charts, it’s used for profiling the customer base. If shopper X loves Kerri Chandler, it’s quite likely that he/she will like releases on the label Freerange. If shopper Y bought 3 tracks by Disclosure it’s very likely they will be interested in a mailshot about the new album when it drops. You get the picture? Now imagine if you improved the sniffing algorithm by incorporating something like Melodyne’s DNA tech, enabling it to pick out melodies, basslines, riffs etc and map them to MIDI. Algorithms could build a ‘structure map’ of every single track out there. Did the bassline come in at 32 bars? Was there a pre-chorus? A middle 8? All of these questions would be answered by the algorithm and put into a database. Combine this with additional data data sources; the streaming sites and the sales sites and also just for fun throw in a little social data such as facebook likes, shares, youtube insights data and you will be facing the SKYNET of music composition, something that could threaten every composer/producer on the planet.
Your laptop/home computer is unlikely to have enough storage space or processing power to do this. A distributed processing network spanning the web that you access from within Ableton/Cubase/Logic etc could be the method used. You and thousands of others will share the load in computing the best choice of instruments to use in combination with each other, the best time to drop that bassline, the best musical key to start with, the perfect length of the track for YOUR favourite DJs. The list goes on, every single possible permutation will have been analysed and the best of the list presented to you, with case examples from historic releases (which you can buy of course) Sliders will allow you to mix up influences; existing track a bit too Goldie? want a touch of LTJ bukem? Just move your sliders on the interface, dial the flavours up. And how about this? these sliders might be monetised. If you dial in a bit of an artist they will get paid for it 😉 It’s all gonna get crazy over the next couple of decades people… Strap yourself in, it’s gonna be a wild ride!
PS if you build something using my ideas please make sure you cut me in on the profits. THANKS!