Curated and human annotated (singing quality score) subset of DAMP singing vocals dataset (no audio files) For obtaining the corresponding audio files, please visit https://ccrma.stanford.edu/damp/.
https://drive.google.com/file/d/1pDQrrWog3dJbl1olTImgdTw_flkWj6ST/view?usp=sharing
This dataset is a curated subset of Smule's DAMP dataset (without audio files). It consists of 4 songs, each sung by 100 singers, so a total of 400 singing renditions. The songs were evaluated for their singing quality through a crowd-sourcing platform, where measures were taken to ensure quality of rating. A detailed description of the way this dataset was curated and the human annotation collection procedure are discussed in the following paper. Please cite this paper if you happen to use this dataset for your research. Thanks!
C. Gupta, H. Li, and Y. Wang, "Automatic leaderboard: Evaluation of Singing Quality without a Standard Reference," IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 28, pp. 13–26, 2020.
https://ieeexplore.ieee.org/document/8871113
For any queries, contact: chitralekha@u.nus.edu
This dataset is purely meant for research purposes, and must not be used for commercial purposes.