Content fingerprinting and matching

Design a content-matching / copyright-ID system (Content-ID style) that fingerprints every uploaded video and audio track and matches it against a reference catalog of tens of millions of copyrighted works. Every new upload (hundreds of hours per minute) must be checked, partial and modified matches (pitch-shifted audio, cropped/recompressed video, a 10s clip inside a 30m vlog) must be caught, and a match must trigger a claim. Design the fingerprinting, the index, and the matching pipeline.