There are two ways to recognise the same song in two different files, and they have very different properties. Being clear about which one is in use tells you exactly when to trust the results.
Acoustic fingerprinting decodes the audio and derives a signature from the sound itself — the approach music-identification services use. It works on completely untagged files and recognises a track regardless of what the metadata claims. It is also computationally heavy, needs a decoder for every format, and is the kind of subsystem that brings its own licensing considerations.
Tag matching reads the metadata the file already carries. It is fast, needs no decoding, and is exactly as reliable as your tags are.
DupeDoctor does tag matching. It reads title, artist, album, and duration from each audio file and forms a grouping key from the lowercased artist and title plus the duration in whole seconds. Files with neither an artist nor a title are skipped rather than guessed at.
The duration component is doing more work than it looks like. Without it, every track sharing an artist and title would group together — remasters, live versions, radio edits, and re-recordings all collide. Requiring an exact whole-second match separates them, since different recordings almost never run to the same second. The engine has a test covering exactly this: two tracks with the same artist and title at 200 and 245 seconds do not group.
What this means in practice. On a well-tagged library — anything organised by a music manager, ripped with a tagger, or bought from a store — matching is accurate and fast. On a library of files named `track01.mp3` with empty tags, song matching will find nothing, and that is the honest outcome rather than a bug. Those files are still covered by exact matching if any of them are byte-identical copies.
One caveat on precision. Duration is compared as whole seconds, so the same song encoded by two different encoders can occasionally differ by a fraction that rounds either way and lands the copies in separate groups. That is a conservative failure — it under-groups rather than over-groups, which is the right direction to fail in for a tool that deletes files.
Audio groups, like photo and video groups, are never selected automatically. They also do not contribute to the reclaimable headline, which counts only exact and copy-named matches.
