New Database Allows Users to Explore Music Used for AI Training

A recent discovery by a reporter has revealed four extensive music datasets that are being utilized to train AI models, now made fully searchable for everyone. Among these, two datasets stand out with a staggering 12 million and 9 million tracks, while the other two, though smaller, still contribute significantly with over 100,000 songs each.

These datasets have been downloaded thousands of times, and while the exact users remain unknown, both Google and Stability have acknowledged their use in research papers. Some sources, like the Free Music Archive dataset, allow free streaming for personal use but require licensing for any commercial endeavors.

However, accessing these datasets isn’t as straightforward as simply downloading a ZIP file and feeding it to an AI model. As noted, three of the datasets consist of links to songs on platforms like YouTube or Spotify. AI developers often resort to tools that automate the downloading process, which can bypass logins, ads, and other mechanisms that support content creators. This practice breaches the terms of service of these platforms.

The dataset features a wide array of artists, from pop icons like Lady Gaga and Fred Again.. to legendary acts such as Radiohead, Aphex Twin, Wu-Tang Clan, Bruce Springsteen, and avant-garde composer Hainbach. You can explore the songs, books, and other media being used to train AI models through the dedicated AI Watchdog site.