Abstract

Identification and extraction of singing voice from within musical mixtures is a key challenge in source separation and machine audition. Recently, deep neural networks (DNN) have been used to estimate ‘ideal’ binary masks for carefully controlled cocktail party speech separation problems. However, it is not yet known whether these methods are capable of generalizing to the discrimination of voice and non-voice in the context of musical mixtures. Here, we trained a convolutional DNN (of around a billion parameters) to provide probabilistic estimates of the ideal binary mask for separation of vocal sounds from real-world musical mixtures. We contrast our DNN results with more traditional linear methods. Our approach may be useful for automatic removal of vocal sounds from musical mixtures for ‘karaoke’ type applications.

Bibtex 


@inbook{Simpson_2015,
  author = {Simpson, A. J. R. and Roma, G. and Plumbley, M. D.},
  editor = {Vincent, Emmanuel and Yeredor, Arie and Koldovsk{\'y}, Zbyn{\v{e}}k and Tichavsk{\'y}, Petr},
  title = {Deep Karaoke: Extracting Vocals from Musical Mixtures Using a Convolutional Deep Neural Network},
  booktitle = {Latent Variable Analysis and Signal Separation: 12th International Conference, LVA/ICA 2015, Liberec, Czech Republic, August 25-28, 2015, Proceedings},
  year = {2015},
  publisher = {Springer International Publishing},
  address = {Cham},
  pages = {429--436},
  isbn = {978-3-319-22482-4},
  doi = {10.1007/978-3-319-22482-4_50},
  keywords = {"maruss"}
}

Implementation

Source code for the experiments described in this article is available for download. The code runs on Matlab 2014. The dataset used for evaluation is a subset of medleydb. The code includes a json file with the index of the subset, you still need to obtain a copy of the dataset from the original authors.

Download (beta)

The software is distributed under BSD license. If you use it, please cite us.

References

R. B. Palm, “Prediction as a candidate for learning deep hierarchical models of data,” Master's thesis, Technical University of Denmark, Denmark, 2012.

“MedleyDB: A Multitrack Dataset for Annotation-Intensive MIR Research,” in Proceedings of the 15th International Society for Music Information Retrieval Conference, Taipei, Taiwan, 2014.