Deezer
In-house detection of fully AI-generated music
Deezer built a tool to detect fully AI-generated tracks and keep them out of recommendations
Problem
Music generators such as Suno and Udio let anyone produce tracks in bulk. Deezer received about 10,000 fully AI-generated tracks a day in January 2025, around 10% of its daily deliveries, and some of the streams on such tracks were fraudulent.
Approach
Deezer developed its own detection tool, which finds the artifacts that generation models leave in audio and can help identify which model made a track. It does not need a separate training set for each model. Deezer filed two patent applications in December 2024. The tool supports a tagging system that labels fully AI-generated music and excludes it from algorithmic and editorial recommendations.
Results
Takeaway
A streaming service treated AI-generated uploads as a content-quality and fraud problem and built its own filter. The accuracy figure is Deezer's own.
Categories
- Industry
- Media and entertainment
- Business function
- Production and quality
- Innovation type
- Process innovationProduct and service innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Built in-house
- Goal
- Quality and risk
- Evidence
- Company's own claim
Sources
Source summaries
Deezer's announcement describes its in-house tool for detecting fully AI-generated music and the planned tagging and recommendation exclusion. It reports about 10,000 such tracks a day at launch. Later updates on the page add a 99.8% accuracy claim.
Full details are in the original sources.