Training neural audio classifiers with few data

TitleTraining neural audio classifiers with few data
Publication TypeConference Paper
Year of Publication2019
Conference Name44th IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP2019)
AuthorsPons, J., Serrà J., & Serra X.
Conference LocationBrighton, UK
AbstractWe investigate supervised learning strategies that improve the training of neural network audio classifiers on small annotated collections. In particular, we study whether (i) a naive regularization of the solution space, (ii) prototypical networks, (iii) transfer learning, or (iv) their combination, can foster deep learning models to better leverage a small amount of training examples. To this end, we evaluate (i-iv) for the tasks of acoustic event recognition and acoustic scene classification, considering from 1 to 100 labeled examples per class. Results indicate that transfer learning is a powerful strategy in such scenarios, but prototypical networks show promising results when one does not count with external or validation data.
preprint/postprint documenthttps://arxiv.org/abs/1810.10274
intranet