Music Identification and Monitoring
Identifying a known recording is a different problem from understanding music. It is the problem of rights and of monitoring: which song is playing, in this bar, through this microphone, over this crowd. And it is a problem where laboratory accuracy and deployed accuracy part company faster than almost anywhere else in audio.
Evaluate the way it will be used
Our work on audio fingerprinting combines contrastive and transfer learning, but the part we care most about is the real-world evaluation protocol: measuring identification through a microphone, in rooms, at distance, against noise and crowd, rather than on clean queries. We release the benchmark — deep architectures, training pipelines and database indexing techniques — so that the comparison can be repeated.
Where this runs
Museek developed a lightweight, device-friendly system for automatically monitoring and recognising music played in public venues — restaurants, hotels, cafes, clubs, bars — in collaboration with GEA, the official music copyright collective management organisation of Greece.
The same capability is delivered as a service in two European platforms: music fingerprinting in DeployAI, the European AI-on-Demand Platform, and song identification for monitoring music use in broadcast and public spaces in the PHAROS AI Factory.
Selected publications
- Contrastive and Transfer Learning for Effective Audio Fingerprinting through a Real-World Evaluation Protocol IJMSTA, 7(1), 68-82 (2025) doi
Open resources: deep-audio-fingerprinting-benchmark, an evaluation benchmark for song identification systems covering deep architectures, training pipelines and database indexing techniques.
If you run a venue network, a broadcaster or a rights organisation and need identification measured under your own conditions rather than a vendor’s, that is a conversation we have had before.