ChoralSynth is a synthesized dataset of 20 multitrack choral songs curated by carefully listening and analyzing a set of synthetic choral songs generated using one of the state-of-the-art synthesizers. The resulting dataset can serve as a valuable resource for various MIR research endeavors like source separation, melodic analysis, chord analysis, rhythmic analysis amongst others.
- Manual Criteria Design: The initial step involved the manual design of filtering criteria. This design was based on a careful examination of over 300 scores and their corresponding synthesized versions to ensure comprehensive and relevant criteria. \
- Automatic Filtering: Subsequently, the designed criteria were used for automatic filtering of the scores. \
- Manual Verification: The filtered songs from step 2 were then examined again manually, validating the synthetic versions through attentive listening, ensuring they demonstrated completeness and a resemblance to human singing.
Detailed statistics and insights regarding the automatic filtering process, along with resulting statistics, can be found in the accompanying supporting notebook. Additionally, CSV files are provided, indicating the specific scores that were selected for further analysis.
At this point, we have curated a set of 20 songs, however, the methodology can be extended to add more songs from the resulting list.
The dataset is available for conducting non-commercial research related to choral singing. It is available for download on Zenodo as well.
The authors recommend the use of virtual environments.
git clone https://github.com/MTG/ChoralSynth.git
cd ChoralSynth
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
.
├── Dataset
├── README.md
├── choralsynth.txt
├── data_analysis_viviana.xlsx
├── requirements.txt
└── src
└── scripts
├── CPDL data filtering.ipynb
├── Convert to MIDI.ipynb
├── correct_file_names_removed_treble_wrong_lyrics_cpdl_repeat1.csv
└── correct_file_names_removed_treble_wrong_lyrics_cpdl_repeat2.csv
Please use the following publication when using the dataset:
Narang, J., De La Vega, V., Lizarraga, X., Mayor, O., Parra, H., Janer, J., & Serra, X. (2023). ChoralSynth: Synthetic Dataset of Choral Singing. arXiv preprint arXiv:2311.08350.
Bibtex version:
@article{narang2023choralsynth,
title={ChoralSynth: Synthetic Dataset of Choral Singing},
author={Jyoti Narang and Viviana De La Vega and Xavier Lizarraga and Oscar Mayor and Hector Parra
and Jordi Janer and Xavier Serra},
year={2023},
eprint={2311.08350},
archivePrefix={arXiv},
primaryClass={cs.SD}
}
ChoralSynth is licensed under CC BY-NC-SA 4.0
