Jon Stark (France Médias Monde)
Manual copyright declarations for music are time consuming for those undertaking the task and the results are often inaccurate. There are difficulties too for the authors’ rights societies, related to half-yearly deliveries, non- compliant formats, lack of metadata, etc.
Over several years, algorithmic companies specializing in music recognition have emerged, but their collaborations with the various media providers are often fragmented and limited. France Médias Monde (FMM) has carried out real research and innovation work on this subject, even having a staff member dedicated to addressing the challenge.
Using fingerprinting
The implementation of audio fingerprinting within the company started from scratch. As the technology is based on Shazam-type music recognition, it was necessary to test the various solutions of companies specializing in fingerprinting to identify the most relevant and robust. A trusting relationship would be essential to the good running of the project.
A precise and detailed technical survey was carried out to understand how music is produced, used, broadcast, and declared. It was also necessary to identify the most relevant metadata and the requirements of the authors’ rights societies.
After several phases of testing and correction, the technology has made it possible to detect more elements, particularly background music. We were able to obtain more than 90% music recognition on MCD and France 24 (see box).
Automated declarations
Once fingerprinting was implemented, the declarations for the four language variations of France 24 were automated. These declarations contain all the essential data related to the music broadcast (title, authors, duration, date, ISRC) as well as other information such as programme names or the music type, allowing its contextualization.
These statements were enthusiastically welcomed by SACEM (the French copyright management society), which now automatically receives accurate data every month. This process allows for better payment of music royalties. In general, the use of this technology allows real productivity gains for all the production and management teams of individual channels.
The main limitation of this technology is for live performances in programmes. It is only possible to identify such live broadcasts after the event, once the extract concerned has been manually tagged in the database.
Additional data
Fingerprinting makes it possible to extract a large amount of music data from the reports for the calculation of the proportion of on-air time where music is used. We can specify the nature of the music data (e.g., proportion of music idents, ranking of the most-played music, by period).
By extending the fingerprinting, we can also provide RFI’s advertising department and our advertisers with real, time- stamped reports on the broadcast of their spots on our stations. In addition, this data allows us to automatically calculate the proportion of the type of advertising broadcast (sponsorships, station promos, advertising) on our channels to verify that quotas agreed with the French regulatory authority are respected.
Future plans include extending the technology to automated declaration of news reports and other spoken content broadcast on RFI. Finally, FMM programmes published on third-party platforms (podcasts, YouTube videos, etc.) may eventually be covered by the automatic declaration.
This article was first published in issue 55 of tech-i magazine.
Photo credit: Robin Cussenot