Technology

Fingerprints that survive the remix.

Our engine identifies your recordings even when they are sped up, pitched, re-recorded, clipped or regenerated by AI - and scores every match, so humans decide the hard cases.

Book a demo How it works
Matching

How a song becomes a fingerprint.

Spectral landmarks

Sound, reduced to structure

Each recording is distilled into a compact constellation of time-frequency landmarks - a fingerprint that identifies the audio without needing to store or ship the audio itself.

Segment-level matching

Clips, not just full tracks

We match segments, not whole files. A short clip buried inside a long video or livestream can be located with exact in and out points - the timestamps that go into the evidence pack.

Manipulation-aware

Built for evasion edits

Matching is designed to withstand the usual tricks - speed and pitch shifts, EQ, re-encoding, cropping - so a "nightcore" upload or slowed remix still points back to your master.

Detection classes

What the engine flags.

Every flag carries a confidence score. High-confidence exact matches move fast; harder classes route to analysts.

Masters

1:1 uses

Direct re-uploads and embedded uses of the original recording - the clearest class, matched acoustically against your fingerprints.

Covers & re-records

Same song, new recording

Melodic and harmonic matching surfaces covers and re-records - a composition-side right, enforced differently from masters.

Derivatives

Remixes, edits, loops

Sped-up and slowed versions, remixes, loops and sampled fragments - matched at segment level even when heavily transformed.

AI versions

Clones & regenerations

AI voice clones and model regenerations of your works are flagged for analyst review - the fastest-growing and most contested class, which is exactly why a human decides.

Two rights, one scan

Recording and composition are different rights.

The master (the recording) is matched acoustically. The composition (the song itself) is matched melodically - that is how a cover with none of your original audio still shows up.

The distinction drives enforcement: a re-upload of your master can be taken down almost anywhere, while a cover is often a licensing and monetization question rather than a removal. Our analysts route each case down the correct path. More in our Resources library.

Case routing
Master match · acoustic fingerprinttakedown / claim
Cover detected · melodic matchlicense / monetize
Derivative · transformed segmentreview → action
AI version · clone or regenerationanalyst review
Verification

Machines rank. People decide.

No claim leaves SONPROOF on a machine's word alone. Analysts confirm the match, check for licenses and obvious fair-use contexts, and assemble the evidence pack before a case ever reaches your approval queue.

This is a legal requirement as much as a quality bar: takedown regimes like the DMCA penalize misrepresentation. Verified evidence and your recorded authorization protect both of us.

  • Confidence thresholds tuned per detection class - exact masters clear quickly, ambiguous derivatives get human time.
  • Known licensees and authorized channels are excluded before review, not after.
  • Borderline and AI cases always route to an analyst - never auto-filed.
  • Every verified match becomes a sealed evidence record: segment refs, score, source, reviewer and timestamps.
Next step

Test it against your hardest catalog.

Bring the releases that get remixed, cloned and re-uploaded the most - and see what the engine finds.