More Bach than Beatles?
Deep learning algorithms are singing new tunes as AI now generates a wide range of musical genres
In the year 2007, as a part of their album ‘Music Heads’, Israeli band Radio Trip had released an electronic track entitled ‘Computers Singing’. It was more of an eclectic dance song than a premonition for the future. Perhaps they’ll be glad to know that it worked both ways.
Computer-generated music has advanced so much over the past few years, that it is no longer a startling novelty. It has, in fact, become a staple in the music industry, being used by several composers in their creative processes. Back in the 1990s, David Bowie famously assisted in the development of a software called ‘Verbasizer’ – which could source existing literary material and, using a combinatorial sequencing on the words, churn out fresh lyrics in minutes. It is prudent to note in this regard that, for AI-based music generation systems, it is much easier to produce orchestral classical or ambient music rather than unpredictable pop-tunes.
This is due to the regular schematic mathematical structures present in most classical music, which can easily find symbolic representation in AI-based systems. For AI, it is thus much easier to produce Bach than Beatles.
But we’ve come a long way since then. In 2016, Sony used a software called ‘Flow Machines’ to develop a melody akin to a Beatles tune – and in the hands of (human) music composer Benoît Carré it was transformed into a full-fledged pop track – ‘Daddy’s Car’ – which, we think, sounds rather lovely. Aiva Technologies, too, has been on the forefront of AI-generated rock music for several years now.
In a fresh development in the world of manufactured pop, OpenAI has recently launched a neural network called Jukebox – which serves as a repository for all musical genres.
With a database ranging from teenybop and country to hip-hop, grunge and heavy metal, the algorithm can develop musical tracks in almost any genre, autocomplete songs once you give it the first few seconds, and even ‘sing’. It accepts lyrics, genres and artists as inputs to produce passable imitations of famous music styles. It’s basically the newest Elvis impersonator on the block.
