VOICe — mirror

This is a mirror, not the official release. The audio is redistributed for reproducibility of our experiments, with credit to the original authors. Use the original record and cite it.

Original dataset

• VOICe: a dataset for generalizable sound event detection and domain adaptation.
• Authors: Shayan Gharib, Konstantinos Drossos, Eemi Fagerlund, Tuomas Virtanen.
• Zenodo: https://zenodo.org/records/3514950 (DOI 10.5281/zenodo.3514950).
• Contents: 1,449 mixtures of three sound events (baby crying, glass breaking, gunshot) mixed with background noise from several acoustic scenes at controlled SNRs.

What is included

Extracted from the official archives, not re-encoded:

Folder	Source archive	SNR
clean/	VOICe_clean.7z	clean
minus9/	VOICe_snr_-9dB.7z	−9 dB

The −3 dB archive is not mirrored here. Metadata and file layout are preserved as in the original archives.

Licence and redistribution note

• Original licence: other-at, as listed on the Zenodo record. Follow the original terms; this mirror grants no broader rights.
• Research use only. No authorship of the recordings is claimed.
• If the original authors object, the repository will be taken down.

Citation

Cite the original VOICe record (DOI 10.5281/zenodo.3514950) and the associated paper by Gharib et al.