MedleyVox: An Evaluation Dataset for Multiple Singing Voices Separation

<p>MedleyVox is an evaluation dataset for multiple singing voices separation, which consists of&nbsp;381 segments (1.1hour),&nbsp;containing 23 songs from MedleyDB v1 and v2 (https://medleydb.weebly.com).</p> <p>&nbsp;</p> <p>MedleyVox&nbsp;contains 1) unison, 2) duet, 3) main vs. rest (folder name &#39;rest&#39;) , and 4) N-singing (&#39;unison&#39; + &#39;duet&#39; + &#39;rest&#39;, please check our dataloader code&nbsp;for N-singing in&nbsp;https://github.com/jeonchangbin49/MedleyVox/blob/main/svs/data/test_dataset.py)&nbsp;categories explained in our ICASSP 2023 paper. For more details, please check our paper (https://arxiv.org/pdf/2211.07302.pdf) and code repository (https://github.com/jeonchangbin49/MedleyVox).</p> <p>&nbsp;</p> <p>Acknowledgment</p> <p>We are grateful to Rachel Bittner, the author of the original MedleyDB data, for allowing us to publish the&nbsp;MedleyVox&nbsp;dataset.</p> <p>&nbsp;</p> <p>License</p> <p>This work is licensed under a&nbsp;<a href="https://creativecommons.org/licenses/by-nc-sa/4.0/">Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License</a>.</p>