Spotify launches AI music tool, Basic Pitch


Spotify has launched an open-source AI-powered tool called Basic Pitch. It’s an audio-to-MIDI converter: a user can upload an audio recording of an instrument, or a voice, and the tool uses machine learning via a neural network to transcribe that input into MIDI – the musical digital standard used by almost all electronic musical instruments. It means that your idly-plucked ukulele melody can then be used in a DAW like Garageband and be adapted or tweaked to “play” any digital instrument. It’s “built for artists and producers who want an easy way to turn their recorded ideas into MIDI,” says Spotify, whose Audio Intelligence Lab collaborated with Spotify-owned Soundtrap to create the tool.

Music Ally tried it out, la-la-la-ing a simple series of notes, and the website quickly returned a faithful replica of it in MIDI notation, including all the wobbly pitch bends this tuneless singer provided. The website even played it back as a virtual guitar-like instrument: it was impressive to hear a voice instantly played back as something else entirely. (You can try it here.)

The website makes it clear that none of your audio is saved or stored –  it all stays on the user’s machine. Presumably, however, the Machine that converts that audio into notation is also Learning about how musicians write music, and which notes human beings think go where when they are making it. This information could perhaps be used in the future for all sorts of AI-powered songwriting applications, including music that is eventually streamed on Spotify.

“We can imagine how the model could be integrated into real-time systems — for example, allowing a live performance to be automatically accompanied by other MIDI instruments that “react” to what the performer is playing,” suggests Spotify. The tool is open-sourced, meaning that if you are tech-minded, you can download the project’s Python library and integrate it into your own audio tools.

Written by: Joe Sparrow