Machine Learning Unlocks a New Era for Music
Where the Internet Lives
00:00
How to Use Machine Learning to Improve Musical Performance
Using big machine learning models, there's just too many computations to do this. The researchers on magenta developed an approach called differential ball digital signal processing that allowed them to control efficient signal processing operations with expressive neural networks. Make neural audio synthesis orders of magnitude faster and more data efficient by baking prior knowledge into the network. What that means is you can have machine learning models that are a lot smaller. You could generate something and get it back in a couple of seconds. This was a game changer because it put music machine learning in the audio space into an interactive domain which is very important for artistic and musical applications.
Transcript
Play full episode
