A folder of track01.mp3 isn't a metadata problem — it's a content problem. The audio knows the song. Sound Recognition listens and renames.
Every DJ library accumulates files that have no useful identity. Tag readers see them as `track01.mp3` or `unknown.flac`. Filename searches return nothing. Manual identification means listening to each one and hoping you recognise it — a process that scales to hundreds of files only on the patience of someone who has nothing else to do on a Sunday. The audio waveform itself contains the song's identity; the metadata around it just lost the memo somewhere between the original encoder and the USB transfer that lost the tags. The right tool restores the connection.
Supported today
Rekordbox · Serato DJ · VirtualDJ (incl. Favorite Folders) on Windows 10+ and macOS (Apple Silicon + Intel).
The architectural choice that makes Sound Recognition work cleanly is treating the audio as the source of truth. The filename, the tag, and the DJ database are downstream consequences of "what song is this" — once that question is answered, propagating the answer to every layer is straightforward. Tools that try to fix mis-named files via metadata heuristics (fuzzy-matching the broken filename against a track list, say) cap out at "better than nothing" because they can't recognise a file whose filename is completely meaningless. Sound Recognition recognises what's in the file, not what's around it. And the server-side proxy + per-license quota architecture means the recognition service stays a sustainable cost rather than an open spigot.
The recognition service is a paid third-party API. MLD covers the per-recognition cost from the Pro license fee, with a daily quota that's generous for normal cleanup use but bounded enough that the cost stays predictable. Putting it behind the Pro gate keeps the quota meaningful for users who actually need it.
No. A short audio fingerprint (a compact hash, not the audio file) is sent through MLD's server-side proxy to the recognition provider. The audio file stays on your machine. The proxy is MLD's own infrastructure, not the third party's.
MLD shows you the score and lets you decide. Below 70% confidence is flagged for review. You can listen to the file in the built-in player and confirm by ear, or reject and try a different recognition by sampling a different segment of the audio.
Recognition is only as good as the reference database. Released music covered by the major catalogs (which is most commercially distributed music) gets identified reliably. Promos that have never been released, custom edits, and mashups may not be recognisable — those are the edge cases where the technology can't help.
Both use audio fingerprinting, but different jobs. The duplicate scan finds files in your library that contain the same recording — local fingerprint match. Sound Recognition matches a fingerprint against the world's music database to identify what the song actually is. Same underlying technique, opposite directions.
Yes. MLD takes a timestamped backup of your Rekordbox master.db / Serato crates / VirtualDJ database before any rename operation. Restore is one click if the rename ever looks wrong.
Yes. Point Sound Recognition at a folder of unknown files and it recognises each one, queues the rename proposals, and lets you approve them in bulk with per-file accept/reject for borderline matches.
Free tier covers detection and viewing forever. Pro is $49 lifetime — paid once, no subscription.
