For external participants: Please join our mailing list to receive announcements of future C4DM seminars: http://c4dm.eecs.qmul.ac.uk/seminars.html
Date and Time Monday, 17th July 2017, at 4:00pm
Place Most rooms are booked for graduation. Therefore, the talk will be in:
Room Law 1.00, Laws Building, Queen Mary University of London, Mile End Road, London E1 4NS. Information on how to access the school can be found at http://www.eecs.qmul.ac.uk/contact-us/.
Speaker Ching-Wei Chen and Andreas Jansson
Abstract Ching-Wei will begin the talk with an introduction of Spotify's music personalization features, such as Discover Weekly and Release Radar, with a brief mention of some Machine Learning and Audio Analysis technologies being used behind the scenes. Andreas will then continue with a presentation of his recent work in vocal separation:
The decomposition of a music audio signal into its vocal and backing track components is analogous to image-to-image translation, where a mixed spectrogram is trans-formed into its constituent sources. We propose a novel application of the U-Net architecture — initially developed for medical imaging — for the task of source separation, given its proven capacity for recreating the fine, low-level detail required for high-quality audio reproduction. The model is trained on a large set of data automatically derived from the Spotify commercial music catalogue.
Bio Ching-Wei is a Chapter Lead for Machine Learning at Spotify in NYC, where members of his team work on music recommendation and analysis. Previously he was at SoundCloud managing their Content ID/Copyright team, and before that at Gracenote where he researched musical mood and tempo classification.
Andreas works as an engineer / researcher in the music understanding group at Spotify in NYC. Before that he was a developer at The Echo Nest in Boston and This Is My Jam in London. He's also a part-time PhD student at City University, supervised by Tillman Weyde.