Deep learning | Jordi Pons

Preprint: “GASS – Generalizing Audio Source Separation with Large-scale Data”

By Jordi Pons in Conferences, Deep learning October 15, 2023

This paper summarises the latest work I did at Dolby. We study a single general audio source separation (GASS) model trained to separate speech, music, and sound events in a supervised fashion with a large-scale dataset.

Check it on arXiv and listen to our demos!

On Prompting Stable Audio

By Jordi Pons in Deep learning, Music, Teaching September 13, 2023

Stable Audio allows you creating custom-length audio just by describing it. It is powered by a generative audio model based on diffusion. You can generate and download audio in 44.1 kHz stereo. You also have a nice interface, no need to be a hacker! And the audio you create can be used in your commercial projects. I’ve been experimenting with it during the last weeks, and here some ideas on how to use it!

5 ideas from EUSIPCO 2023

By Jordi Pons in Conferences, Deep learning September 8, 2023

This year I was at EUSIPCO to present the paper that wraps up our work on upsampling artifacts. In the following website you have a summary of our work, including a video “for dummies” with the main ideas:

http://jordipons.me/apps/upsamplers/

ISMIR 2023 paper: “Mono-to-stereo through parametric stereo generation”

By Jordi Pons in Conferences, Deep learning August 1, 2023

In this work lead by Joan, we explore the use of generative models that operate on top of a parametric stereo domain to generate plausible stereo samples from mono audio signals.

Check it on arXiv!

WASPAA 2023 paper: “CLIPSonic – Text-to-Audio Synthesis with Unlabeled Videos and Pretrained Language-Vision Models”

By Jordi Pons in Conferences, Deep learning July 1, 2023

This paper is the result of Herman’s internship at Dolby. In this work we explore the feasibility of training a text-to-audio synthesis model without text-to-audio pairs.

Check it out on arXiv and listen to our demo!