Monday, February 24, 2014

Thought Experiment: Infinite Spotify Playlists Via Procedurally-Generated Tracks



I use Spotify to listen to a lot of instrumental music when working; as it is less distracting than music with vocals. The Spotify Radio feature generates an effectively nigh-infinite playlist of tracks based on the characteristics of a chosen source genre, artist or track.

What appeals to me about this technology is that is encourages a very different mindset from that fostered by the old economic models. I actually like the idea of hearing lots of novel music without knowing who the artists are and quite likely never hearing that one particular piece of music ever again. It can be a river of sound: never the same experience twice.

However, how comprehensive are those playlists today? At what point would the artists or songs start to repeat themselves too much?

It strikes me that one factor which makes Spotify Radio so compelling is that it allows you to hone in on your own favourite sub-niches of sub-niches of sub-niches within musical genres. But at a certain point surely any personalised niche becomes a finite set. Perhaps it is defined by a set period of geography or of time: no-one makes music of that sort any more. Or perhaps it is defined by a small set of artists, or other such variables.

Could we reach a point in the future where Spotify, or similar services, were able to enhance such playlists with their own procedurally-generated music? Thus giving each user their own ultimately personalised music stream.

What I wonder about is the state-of-the-art regarding computer-generated music. Has the technology advanced to the point such that an algorithm could take a detailed blueprint of the DNA of a particular musical sub-genre (such as Pandora’s Music Genome) and then create novel variations on, and iterations from, those conventions? And, most critically, whether the best results would then rise above the levels of musak? I imagine that this would be more readily achievable for instrumental genres, as including lyrics and vocals would add another layer of complexity.

Before all of the Muso-Purists create a flame war in the comments section about musical integrity and whats real,  lets consider the following. What if for many users the job-to-be-done for Spotify turns out not to be the high-end ideal of “find me great artists whom I can follow and develop an ongoing appreciation for the complete depth of their back catalogue”. Rather what if the job-to-be-done was the more prosaic “find me a large amount of novel music that sounds like the kind of music I prefer”. There is a subtle distinction between those two use-cases. Spotify's marketing emphasises the former (perhaps to keep today's iteration of the music industry on-side) but they could be equally successful serving the second need.

Now perhaps the world’s musical output will always be enough to back-fill people’s infinite playlists. (Given that there already at least one dedicated service for playing only the millions of never-played tracks already on Spotify.) But it is also not beyond the realms of imagination to see such user-customised tracks slipping into playlists at some point in our not-to-distant-future.

This may not be a business which Spotify themselves would wish to enter. If so, then perhaps there is a gap in the market for businesses to start creating such content.


(I am thinking of blogging a few short thought experiments of novel features or blue-sky ideas for products and services that I use regularly.)