Sony's DrumNet tech aims to create drum parts using AI

(Image credit: Future)

First there were drum machines and MIDI, then came audio quantisation and sample replacement, but Sony Computer Science Labs (Sony CSL) has announced that it has created the next potential threat to human drummers in the form of DrumNet: AI tech that uses machine learning to create “musically plausible” drum patterns.

So far, Sony has focussed on using DrumNet to create bass drum patterns - feeding DrumNet a very-nearly rock ‘n’ roll 665 songs spanning a range of genres, featuring rhythm tracks without bass drum parts. 

With this learning in place, DrumNet ‘listens’ to the other audio tracks in your composition, then uses a “16-dimensional vector sampled from independent multivariate Gaussians” to determine the placement and pattern of the drum part in relation to the other instrumentation it ‘hears’.

The results, which can be heard here, demonstrate DrumNet’s abilities, and while the patterns are fairly static examples, it shows DrumNet’s potential for people who require drums in their own music without wanting to spend time programming or using preset MIDI or audio loops.

This isn’t the first time drummers have come up against AI: Logic’s Drummer plugin offers the same ease-of-use with customisable parameters, while Positive Grid’s X Drummer app claims to “Listen to your music as a real drummer would”.

There’s no doubt that these tools make capturing song ideas quick and easy, but for now, it seems like it’s still worth us practicing our Bonham bass drum triplets.

Stuart Williams

I'm a freelance member of the MusicRadar team, specialising in drum news, interviews and reviews. I formerly edited Rhythm and Total Guitar here in the UK and have been playing drums for more than 25 years (my arms are very tired). When I'm not working on the site, I can be found on my electronic kit at home, or gigging and depping in function bands and the odd original project.