Improving music source separation based on deep neural networks through data augmentation and network blending
Authors:
Top Cited Papers
- 1 March 2017
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 261-265
Abstract
This paper deals with the separation of music into individual instrument tracks which is known to be a challenging problem. We describe two different deep neural network architectures for this task, a feed-forward and a recurrent one, and show that each of them yields themselves state-of-the art results on the SiSEC DSD100 dataset. For the recurrent network, we use data augmentation during training and show that even simple separation networks are prone to overfitting if no data augmentation is used. Furthermore, we propose a blending of both neural network systems where we linearly combine their raw outputs and then perform a multi-channel Wiener filter post-processing. This blending scheme yields the best results that have been reported to-date on the SiSEC DSD100 dataset.Keywords
This publication has 25 references indexed in Scilit:
- The 2015 Signal Separation Evaluation Campaign2015
- Deep Karaoke: Extracting Vocals from Musical Mixtures Using a Convolutional Deep Neural Network2015
- Joint Optimization of Masks and Deep Recurrent Neural Networks for Monaural Source SeparationIEEE/ACM Transactions on Audio, Speech, and Language Processing, 2015
- Deep learningNature, 2015
- Deep neural network based instrument extraction from music2015
- Deep NMF for speech separation2015
- Cross-domain cooperative deep stacking network for speech separation2015
- Nonnegative Matrix Factorization with the Itakura-Saito Divergence: With Application to Music AnalysisNeural Computation, 2009
- Lessons from the Netflix prize challengeACM SIGKDD Explorations Newsletter, 2007
- Framewise phoneme classification with bidirectional LSTM and other neural network architecturesNeural Networks, 2005