SingFake

dataset·singing voice deepfake·active

In-the-wild real and deepfake singing data for singing-voice authenticity detection.

Recorded facts

Official site	https://singfake.org/ ↗
Geography	United States
size	hours by partition: documented on project page; exact aggregate not normalized
scope	Singing voice deepfake detection in real-world conditions.
genres	user-generated singing
license	Research-use acknowledgement; no general redistribution license stated
steward	University of Rochester Audio Information Research Lab
creators	Yongyi Zang; You Zhang; Mojtaba Heydari; Zhiyao Duan
languages	multilingual/unspecified
modalities	URL annotations; audio subset; real/fake labels; vocal and mixture variants
geographies	Global
access method	Research-only acknowledgement and URL annotations from official site.
intended uses	singing voice deepfake detection
related papers	SingFake: Singing Voice Deepfake Detection
prohibited uses	Use outside research is not authorized by the reviewed notice.
annotation method	Authenticity labels assigned from source/generation provenance; vocals and mixtures evaluated.
collection method	Collected from popular user-generated content websites and synthetic singing sources.
provenance claims	Project page states only a portion was used in the ICASSP 2024 paper due to download constraints.
current availability	Paper subset and URL annotations available with research-use acknowledgement.
commercial use limits	Research-only use stated.
us market scope basis	Publicly available or materially used in US-facing music/audio-AI research.
limitations biases disputes	URL and download constraints limit reproducibility.; Platform content and languages may be unbalanced.; Authenticity labels can age as generation systems change.