Generative Music

Applied Machine Learning in Arts 🎨

Alexandre DuBreuil

What is generative music?

"Generative art is an artwork partially or completely created by an autonomous system 🤖" - Music Generation with Magenta

But has nothing to do with code!

Lets take our time machine and come back to the 18th century.

In 1792, a guy named Mozard created the Musikalisches Würfelspiel (musical dice) game.

The result of a dice roll determines which measure to add to the generated composition

The system is autonomous, because you don't decide the outcome

That's it, that's generative music

So basically the system is working for (with) you to reach a common goal: great music

But the result is as good as the system that creates it 😬.

Fast forward in time (today!)

Back in 2021, we're sitting on a pile of theorical knowledge about music, and lots of technology.

Generative Music Artists

Brian Eno, who coined the term "generative music", who created numerous infinite albums

Esoteric Languages

ORCA, one of many (esoteric) langages created for generative music.

Electronic Music Scene

Algorave, an underground music scene where the music is live-coded in front the audience (hence its name).

As musicians

We have generative music tools to help us compose, produce, improvise, etc. 🚀

Limitations of algorithms

Climat General (2017)

Claire Malrieux / Alexandre DuBreuil (skip to 11:00)

This generative artwork makes a relation between climate events (rain, tornados, etc.), the artist's drawings, and the music.

How generative can we go?

Generative text, generative voice, generative score (partition), generative audio, generative emotions!? 🤯

Using classical algorithm is HARD

Some things are hard to do using traditional algorithm like text generation, voice generation, audio generation, etc. So we need a way of training a machine on human data...

Machine learning to the rescue!

With machine learning, you can use human data, and let the machine predict / generate based on what it has learned.

DreamBank (2019)

Claire Malrieux / Alexandre DuBreuil (skip to 10:50)

The machine tells us about its dream: "The street of the dream, interpretation too, I was very surprised that first I had..."

DreamBank Generative System

We go "full generative": generative dream text based on human dreams, analysing sentiments (sad, happy, etc.) of the text, generating speech, generating musical score...

Using neural nets for generating partitions

Generating score based on style (rock, jazz, classical, etc.), emotion (sad, happy, etc.), instrument (drum, piano, etc.). With different neural networks (Recurrent Neural Networks, RNNs).

Using neural nets for generating audio

Generating audio is HARD, but we can generate sounds
based on cat meow!

Closing

Thank you!