Lode Vuegen, Peter Karsmakers, Hugo Van Hamme1, Bart Vanrumste1
12:25 - 12:38 | Mon 5 Mar | Treasure Island E | MoBT2.6
This abstract discusses a non-negative matrix deconvolution (NMD) based learning algorithm capable of identifying acoustic events from weakly supervised (WS) data. The weakly supervision is implemented by indicating the events that took place over a longer period of time without identifying beginning nor endings. We will show that WS-NMD is successfully applied for the task of acoustic event classification and that it achieves similar results compared to the baselines on the NAR-dataset but with significant less annotation work.