AI in Audio

Part I: Automated music generation and production
Spring 2019 - Audio Tech Talk Series
March 3, 2019

What is AI?

Terms

Artificial Intelligence

Machine Learning

Neural Networks

Deep learning

Simple Neural Network

Guitar Pedal

"Neural Network" Guitar Pedal

Examples

YOLO

Progressive growing of GANs

Progressive Growing of GANs for Improved Quality, Stability, and Variation | T. Karras et. al. | 2018

What about music?

We have to go way back...

Wind chimes

China - 1100 BC

Aeolian harp

Greece - 6 BC

Markov chains

Iannis Xenakis - Analogique in 1920s

Enter deep learning...

Symbolic music generation with LSTMs

Doug Eck - Blues Improvisation - 2002

Improved symbolic music generation

Sony CSL - In the style of the Beatles - 2016

Raw audio synthesis with WaveNet

Deep mind - Classical piano - 2016

More hybrid and symbolic approaches

Jukedeck

Symbolic + synthesis - 2016

AIVA

In the style Star Wars Theme - 2017

Popgun

Symbolic pop music - 2017

For music production

First automatic mixing system

Dan Dugan - Microphone Mixer - 1975

Grounded theory (Psychoacoustics)

Enrique Perez-Gonzalez - 2009

Knowledge engineering (Mixing rules)

Stuart Mansbridge - 2012

Deep learning is next...

We need more data

End-to-end vs. Parameterized models

Next Talk - March 12

AI in Audio

Part 2: Music Recommendation and Understanding