AI Watchdog: Free Music Archive Dataset

Editor’s note: This search tool is part of The Atlantic’s investigation into how music tracks are taken to train AI tools. This work is part of AI Watchdog, The Atlantic’s ongoing investigation into the generative-AI industry.

A collection of 106,574 music tracks downloaded from the Free Music Archive in 2016. The dataset was assembled by the École Polytechnique Fédérale de Lausanne in Switzerland. Most of the tracks are licensed under Creative Commons licenses that require credit to be given to the artist wherever the track is used, and forbid use in commercial projects. The dataset was used to train generative-AI models by Google, and a 13,874-track subset was used by Stability AI.