in Projects, Programming, Technology

AI for Composing Music

This is part of series on Opportunities for AI in Music Production.

MIDI is the computer format for sheet music, it represents the actual notes and expression of those notes. Some progress has been made in auto-complete for MIDI, but the potential is unrealized. Even random MIDI generation is useful to kick off ideas, but intelligent MIDI generation is missing from DAWs and plugins.

Problem / When composing a song, the writer may not have the skills or inspiration to develop a chord progression or melody. Unfortunately, the “state of the art” is still libraries of pre-canned midi chord progressions and melodies. My favorite tool for this – Scaler – is a great chord builder, and has some new features that make its pre-canned patterns playable, but there’s still a ton of potential.

We’re seeing better and better auto-complete for text, algorithms are writing entire articles now, why not music?

Imagined Solution / Given a key signature, tempo, genre, and possibly some seed notes I’d love a DAW plugin that suggests chords or fills in a melody. It could either just generate notes or even play along with you like a collaborator.

Open AI’s GPT (Generative Pre-trained Transformer) work seems the most promising for this. The approach would be to train on a bunch of MIDI, then use the GPT algorithm to generate more. The bulk of their efforts are focussed on generating text, but some impressive work has been applied to music. Musenet is the most promising, “a deep neural network that can generate 4-minute musical compositions with 10 different instruments and can combine styles from country to Mozart to the Beatles

GarageBand has excellent AI drummers that write their own fills and can be given direction in terms of the complexity and intensity of the pattern they are playing. That’s the interactive experience we should aspire to, it should feel like you’re collaborating with someone else.

Next: AI for Music Arrangement

Write a Comment