Data augmentation for instrument classification robust to audio effects

TitleData augmentation for instrument classification robust to audio effects
Publication TypeConference Paper
Year of Publication2019
Conference Name22nd International Conference on Digital Audio Effects (DAFx-19)
AuthorsRamires, A., & Serra X.
Conference Start Date02/09/2019
Conference LocationBirmingham, United Kingdom
KeywordsAudio Effects, Automatic Instrument Classification, convolutional neural networks, Data augmentation
AbstractThe repurposing of audio material, also known as sampling, has been a key component in Electronic Music Production (EMP) since its early days and became a practice which had a major influence in a large variety of musical genres. Automatically classifying one-shot instrument sounds in unstructured large audio databases provides an intuitive way of navigating them, and a better characterisation. Automatic instrument classification remains an open research topic which has mostly targetted the classification of unprocessed isolated instrumental sounds or detecting predominant instruments in mixed music tracks. For this classification to be useful in large audio databases for EMP, it has to be robust to the audio effects applied to unprocessed sounds. In this paper we evaluate how a state of the art model trained with a large dataset of one-shot instrumental sounds performs when classifying instruments processed with audio effects. In order to evaluate the robustness of the model, we use data augmentation with audio effects and evaluate how each effect influences the classification accuracy.
preprint/postprint document