Most of the current supervised automatic music transcription (AMT) models lack the ability to generalize . The proposed ReconVAT usesreconstruction loss and virtual adversarial training . When combined withexisting U-net models for AMT, it achieves competitive results on common datasets such as MAPS and MusicNet . For example, in the few-shotsetting for the string part version of MusicNet, ReconVat achieves F1-scores of61.0% and 41.6% for the note-wise and note-with-offset-wise metrics . This translates into an improvement of 22.2% and 62.5% compared to the supervised baseline model . Our proposed framework also demonstrates thepotential of continual learning on new data

Author(s) : Kin Wai Cheuk, Dorien Herremans, Li Su

Links : PDF - Abstract

Code :

Keywords : supervised - reconvat - musicnet - note - framework -

Leave a Reply

Your email address will not be published. Required fields are marked *