Starting point
Broadcast and music archives hold decades of material whose metadata stops at "file name and date". Finding a specific take, a speaker, a piece of music or a usable still means listening and scrubbing – by hand.
What we built
Not a media asset management system itself, but three cooperating analysis services that work alongside one, each documented on this site:
- Audio Analyzer – loudness envelope, silence and segment detection, music/speech classification, fingerprint hooks for rights matching
- Video Analyzer – shot and scene detection, keyframes, on-screen text, quality metrics
- Image Analyzer – tagging, faces and objects, duplicates, technical quality
All three are .NET 9 services with REST APIs and queue-based batch processing, so they run next to an existing MAM or archive and write their results back into it as ordinary metadata – no new system to migrate to.
Outcome
Archive searches that were impossible ("all interviews with music under speech longer than two minutes") become queries. Rights reporting starts from detected music instead of manual logs. And because the analyzers are services, the same setup is available for other media companies – as a hosted SaaS or on-premises installation.