Is MixIT Really Unsuitable for Correlated Sources? Exploring MixIT for Unsupervised Pre-training in Music Source Separation
Paper • 2505.07631 • Published
Music Source Separation Models Pre-trained with Mixture Invariant Training (MixIT)
This repository includes pre-trained weights of the following WASPAA 2025 paper. Please see https://github.com/b-sigpro/mixit-mss.git how to use them.
Copyright
Copyright (c) 2025 National Institute of Advanced Industrial Science and Technology (AIST), Japan. All rights reserved.
License
This model is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).
References
If you use any part of this code including the pre-trained models for your work, please cite the paper:
@InProceedings{Saijo2025_MixITMSS,
author = {Saijo, Kohei and Bando, Yoshiaki},
title = {Is MixIT Really Unsuitable for Correlated Sources? Exploring MixIT for Unsupervised Pre-training in Music Source Separation},
booktitle = {Proc. Workshop on Applications of Signal Processing to Audio and Acoustics (WASPAA)},
year = 2025,
month = oct
}