V. Emiya – Audio Inpainting

Carte non disponible

Date/heure
Date(s) - 11/03/2011
14 h 00 min - 15 h 00 min

Catégories Pas de Catégories


We introduce Audio Inpainting, leveraging the concept of Image Inpainting, for the restoration of missing audio data. Audio inpainting covers a wide range of applications – such as click removal, declipping, packet loss concealment or bandwidth extension – where the distorted data is considered as missing and its location is assumed to be known a priori. A framework for audio inpainting in the time domain is proposed, based on sparse representations. It exploits two possible dictionaries (discrete cosine and Gabor) known to provide accurate sparse models for audio signals. The Orthogonal Matching Pursuit algorithm is adapted for audio inpainting, in particular to deal with the properties of the Gabor dictionary. A constrained matching pursuit approach is applied to significantly enhance the performance for the audio declipping problem. Several experiments on real audio data show the global trends and the performance of our approach. Co-work with A. Adler, M. Elad (Technion, Israel); M. Jafari, M. Plumbley (Univ. Queen Mary of London, UK); R. Gribonval (INRIA, Rennes) References: – Part of this presentation is published in “A. Adler, V. Emiya, M. J. Jafari, M. Elad, R. Gribonval, M. D. Plumbley, A Constrained Matching Pursuit Approach to Audio Declipping, ICASSP, 2011.” (see V. Emiya’s web page.) – new publications are planned in 2011.