Data augmentation for deep learning source separation of HipHop songs

TitleData augmentation for deep learning source separation of HipHop songs
Publication TypeConference Paper
Year of Publication2017
Conference Name10th International Workshop on Machine Learning and Music
AuthorsMartel, H., & Miron M.
Conference Start Date06/10/2017
Conference LocationBarcelona
Keywordsdeep learning, hip hop, source separation
AbstractTraining deep learning source separation methods involves computationally intensive procedures relying on large multi-track datasets. In this paper we use data augmentation to improve hip hop source separation using small training datasets. We analyze different training strategies and data augmentation techniques with respect to their generalization capabilities. Moreover, we propose a hip hop multi-track dataset and we implemented a web demo to demonstrate our use scenario. The evaluation is done on a part of the dataset and hip-hop songs from an external dataset.
preprint/postprint documenthttp://hdl.handle.net/10230/32930
Additional material: 
Code available at: https://github.com/MTG/DeepConvSep Hip Hop dataset available at: https://zenodo.org/record/823037 Demo available as: https://hiphopss.github.io
intranet