A Dataset for Automatic Vocal Mode Classification

A dataset for automatic vocal mode classification with vocal modes as introduced by the complete vocal technique (CVT). Data was recorded at the institute for information processing and consists of around 3700 unique productions of sustained vowels. Each production was recorded by up to four microphones, yielding more than 13000 samples in total. For each sample, an annotation, created by three cvt-experienced singers, is provided. Both the individual annotation as well as the aggregated annotation is provided. Details of the creation of the annotation can be found in the corresponding article. For a detailed description, read the readme.txt, which also explains precisely the naming convention and other details.

The dataset contains a few samples that did not show a proper vocal "oscillation" in the folder "ExcludedSamples". They are not part of the actual vocal mode dataset, but were included to perhaps help other research in the future, e.g., to detect singing issues.

If you are using this dataset for your research, please cite (to be published):

@inproceedings{VocalModeDataset,
  title={A Dataset for Automatic Vocal Mode Classification},
  author={Hinrichs, Reemt and Stephan, Sonja and Lange, Alexander and Ostermann, J{\"o}rn},
  booktitle={International Conference on Computational Intelligence in Music, Sound, Art and Design (Part of EvoStar)},
  pages={XXX--YYY},
  year={2026},
  organization={Springer}
}