Title: "Dictionary learning methods for single-channel source separation"
Speaker: Augustin Lefèvre (INMA/ICTEAM, UCL)
Location: "Shannon" Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor
Date / Time (duration): Thursday 29/11, 10h45 (~ 45')
Abstract: We present three main contributions to single-channel source separation. Our ﬁrst contribution is a group-sparsity inducing penalty speciﬁcally tailored for nonnegative matrix factorization (NMF) with the Itakura-Saito divergence: in many music tracks, there are whole intervals where at least one source is inactive. The group-sparsity penalty we propose allows identifying these intervals blindly and learn source speciﬁc dictionaries. As a consequence, those learned dictionaries can be used to do source separation in other parts of the track were several sources are active. These two tasks of identiﬁcation and separation are performed simultaneously in one run of group-sparsity Itakura-Saito NMF.
Our second contribution is an online algorithm for Itakura-Saito NMF that allows learning dictionaries on very large audio tracks. Indeed, the memory complexity of a batch implementation NMF grows linearly with the length of the recordings and becomes prohibitive for signals longer than an hour. In contrast, our online algorithm is able to learn NMF on arbitrarily long signals with limited memory usage.
Our third contribution deals with user informed NMF. In short mixed signals, blind learning becomes very hard and sparsity does not retrieve interpretable dictionaries. Our contribution is very similar in spirit to inpainting. It relies on the empirical fact that, when observing the spectrogram of a mixture signal, an overwhelming proportion of it consists in regions where only one source is active. We describe an extension of NMF to take into account time-frequency localized information on the absence/presence of each source. We also investigate inferring this information with tools from machine learning.
Last updated September 25, 2013, at 09:06 AM