Research Seminars

Low Bitrate Object Coding of Musical Audio Using Bayesian Harmonic Models

Emmanuel Vincent, Queen Mary University of London

Monday 20 March 2006, 5:00pm, Room 105

Abstract

We present a system for low bitrate coding of musical audio that represents mono signals as sets of sound objects composed of harmonic sinusoidal partials. After a brief review of existing methods, we recast this problem in the Bayesian framework. We propose a family of probabilistic signal models combining learnt priors for the object parameters and various perceptually motivated distortion measures for the non-harmonic residual. We design efficient algorithms to infer the object parameters and we build a prototype coder based on the interpolation of the frequency and amplitude parameters. Listening tests suggest that the loudness-based distortion measure outperforms other distortion measures and that our prototype results in a better sound quality than a baseline sinusoidal coder at 8 kbit/s and 2 kbit/s. This work constitutes a new step towards a fully object-based coding system, which would represent audio signals as collections of meaningful note-like sound objects.