https://audio.dev/ -- @audiodevcon​

Deep Learning for DSP Engineers: Challenges and Tricks for Audio AI - Franco Caspe & Andrea Martelloni - ADC23

This talk aims to tackle and demystify the process of the development of an AI-based musical instrument, audio tool or effect. We want to view this process not from the point of view of technical frameworks and technical challenges, but from that of the design process, the knowledge required and the learning curve needed to be productive with AI tools; particularly if one approaches AI from an audio DSP background, which was our situation when we started out.

We are going to quickly survey the current applications of AI for real-time music making, and reflect on the challenges that we found, especially with current learning resources. We will then walk through the process of developing a real-time audio model based on deep learning, from dataset to deployment, highlighting the relevant aspects for those with a DSP background. Finally, we will describe how we applied that process to our own PhD projects, the HITar and the Bessel’s Trick.

Link to Slides:
_

Franco Caspe

I’m an electronic engineer, a maker, hobbyist musician and a PhD Student at the Artificial Intelligence and Music CDT at Queen Mary University of London. I have experience in development of real-time systems for applications such as communication, neural network inference, and DSP. I play guitar and I love sound design, so in my PhD I set out to find ways to bridge the gap that separates acoustic instruments and synthesizers, using AI as an analysis tool for capturing performance features present in the instruments’ audio, and as a generation tool for synthetic sound rendering.
_

Andrea Martelloni

Inventor of the HITar. Interested in applications of deep learning for rich real-time musical interaction and expressive digital musical instruments.
_

Streamed & Edited by Digital Medium Ltd: https://online.digital-medium.co.uk
_

Organized and produced by JUCE: https://juce.com/
_

Special thanks to the ADC23 Team:

Sophie Carus
Derek Heimlich
Andrew Kirk
Bobby Lombardi
Tom Poole
Ralph Richbourg
Jim Roper
Jonathan Roper
Prashant Mishra

#adc #dsp #audio #ai #deeplearning