SIC (Songs ans Instrumentals classification)

Replicable python code for the article:

@unpublished{Bayle2019,
  author = {Bayle, Yann and Hanna, Pierre and Robine, Matthias},
  title = {Content-based information retrieval for supervised classification of musical tags in real-world unbalanced datasets: Application to the Instrumentals and Songs},
  year = {2019},
  note = {Submitted to PLOS ONE}
}

Details

Task: content-based musical playlists generation focused on Songs and Instrumentals.

Musical Dataset: CCMixter, Jamendo, MedleyDB and SATIN.

Results: Our suggested approach generates an Instrumental playlist with up to three times less false positives than state-of-the-art.

Contributions:

• The first review of SIC systems in the context of playlist generation.
• The first formal design of experiment of the Song Instrumental Classification (SIC) task.
• A demonstration that the use of frame features outperforms the use of global track features in the case of SIC and thus diminish the risk of an algorithm being a "Horse".
• A knowledge-based SIC algorithm ---easily explainable--- that can process large musical database whereas state-of-the-art algorithms cannot.
• A new track tagging method based on frame predictions that outperforms the Markov model in terms of accuracy and f-score.
• A demonstration that better playlists related to a tag can be generated when the autotagging algorithm focuses only on this tag.

Requirements/Dependencies

• Python 2 and 3 environments because all scripts works under python 3 except for YAAFE that only works under Python 2
• You need at least 54Go of free space to store the data and audio features
• You need to manually download and store tracks in a dir named: tracks/
• Yaafe's intallation: conda install -c https://conda.anaconda.org/yaafe yaafe
• Marsyas's installation: http://marsyas.info/doc/manual/marsyas-user/Step_002dby_002dstep-building-instructions.html#Step_002dby_002dstep-building-instructions

Alternatives to consider for the SVD and SVS

Singing Voice Detection

• Single-Channel Blind Source Separation for Singing Voice Detection: A Comparative Study
• https://github.com/f0k/ismir2015
• https://github.com/EdwardLin2014/SingingVoiceDetection_Python
• https://github.com/pikrakis/Unsupervised-Singing-Voice-Detection-Using-Dictionary-Learning
• https://github.com/TheaGao/SklearnModel/blob/f6b34cbb88c35a4fc81074fd7f0ab929bf59207a/segmentLabel.py
• https://ieeexplore.ieee.org/document/8334252/
• Revisiting singing voice detection: A quantitative review and the future outlook - Kyungyun Lee, Keunwoo Choi, Juhan Nam
• Singing style investigation by residual siamese convolutional neural networks - Cheng-i Wang and George Tzanetakis, ICASSP 2018
• Singing Voice Detection across Different Music Genres - Scholz, Florian; Vatolkin, Igor; Rudolph, Günter

Singing Voice Separation

• https://github.com/Js-Mim/mss_pytorch
• https://github.com/posenhuang/singingvoiceseparationrpca
• https://github.com/Xiao-Ming/UNet-VocalSeparation-Chainer
• https://github.com/EdwardLin2014/CNN-with-IBM-for-Singing-Voice-Separation
• https://github.com/andabi/music-source-separation
• Musical Instrument Separation on Shift-Invariant Spectrograms via Stochastic Dictionary Learning Sören Schulze, Emily J. King
• [Code][PDF] Music Source Separation Using Stacked Hourglass Networks - Sungheon Park, Taehoon Kim, Kyogu Lee, Nojun Kwak