GEMA launches AI music dataset as licensing debate intensifies

The launch comes nearly two years after GEMA announced plans to develop a framework for licensing music for generative AI training. According to the Berlin-based organisation, PLAI contains almost 180,000 recordings spanning more than 60 genres, with comprehensive metadata and rights cleared for AI training.

The initial music providers participating in the service include Germany-based Bailer Music Publishing, Earmotion Audio Creation, Intervox Production Music, Music Sculptor and Sonoton Music. GEMA said participating rights holders would be compensated proportionately and appropriately, while publishers, collecting societies, authors and other rights providers can join the platform.

GEMA said PLAI offers AI developers a single source for music licensing, with data packages that can be tailored according to factors such as genre and mood. The service is also intended to provide greater legal certainty for companies developing AI tools by ensuring that the music used for training has the necessary rights clearances.

AI-powered music transcription company Klangio is the first announced customer for the dataset. The company’s head, Sebastian Murgul, said the availability of legally cleared training data was important to the business.

“Our customers and we ourselves are part of the music industry,” Murgul said. “That is why fairness and legal certainty are a major issue for us. With the GEMA music dataset, we have found the solution we have been waiting for: legally fully clarified training data where all rights are really covered.”

He added: “This gives us the security to further develop our AI tools on a solid foundation. This shows that fair remuneration and innovative AI development can go hand in hand.”

GEMA director of broadcasting and online, Thomas Theune said the service would simplify access to licensed music for AI developers while creating potential new income for creators.

“Instead of complex rights clearances, AI tool providers now receive high-quality music data including the necessary rights from a single source,” Theune said. “This creates legal certainty for developers and new revenue potential for music creators.”

“With PLAI by GEMA, we show that AI training and fair remuneration of creatives are not a contradiction,” he added. “If both sides are interested in dealing responsibly with each other, we will find joint solutions and enable innovation on a reliable legal basis.”

The launch comes as GEMA continues its legal challenge against AI music company Suno over the use of copyrighted works in AI training. A ruling in the case is expected on 31 July. GEMA has also previously secured a court victory in a dispute involving OpenAI.

GEMA general counsel Kai Welp said the outcome of the Suno case could have wider implications for the licensing of generative AI systems. He called for greater clarity on how creative works used to train AI models should be remunerated, regardless of where the training takes place.

“For the licensing of generative AI providers, we hope for a positive decision on the matter against SUNO on July 31,” Welp said. “It must be made clear that the use of creative works must be remunerated regardless of the location of the training and also when the original works are stored in the systems. This requires the courts, but above all the political framework.”

The launch of PLAI reflects the growing effort by music rights organisations to establish licensing mechanisms for AI training as disputes over copyright, remuneration and the use of creative works continue to escalate. GEMA's model seeks to position licensed datasets as an alternative to potentially unlicensed uses of copyrighted music while offering AI developers a more structured route to obtain training material.