How ryser.id identifies tracks

Every tracklist on ryser.id starts as audio, not as text somebody typed in. This page explains how a set becomes a tracklist, what the badges on a row mean, and where the library sets come from.

From link to tracklist

1. Download and cut. The set is fetched from the link you pasted, converted to audio, and cut into windows of about ten seconds, spread across the whole set. Nothing is kept afterwards: the audio is deleted when the run finishes unless you chose to keep it.
2. Recognise. Each window is sent to audio fingerprinting engines. A full scan uses Shazam, ACRCloud and AudD together; the library scan uses Shazam only. An engine answers with a recording it recognises and a confidence score.
3. Combine. Answers from neighbouring windows are combined into one track with a start and end time. Two engines agreeing raises confidence; one engine alone lowers it. BPM and key are measured from the audio and compared with what the engines claim: a record that plays at the wrong tempo or in the wrong key is demoted, never silently kept.
4. Cross-check. Where a community tracklist for the same set exists, we compare against it to confirm or doubt an identification. We never copy a tracklist from elsewhere; every row here was heard in the audio.
5. Publish. A set you make public, or a library set that names at least 60 percent of its tracks, gets a page with the tracklist, links to each track and artist, and buy links.

What the row labels mean

• Identified: the engines agree and the audio checks pass. This is the normal state.
• Check: something does not add up, for example a tempo that does not match, an artist that does not fit the rest of the set, or a single engine with a low score. The row stays visible so you can judge it; if you know the answer, correct it.
• Unknown: nobody could name this segment. Often this is an unreleased record. We remember the audio, so when the same record turns up in another set the two are linked, and one correction names both.

Library sets

The set library scans the most popular new DJ sets from public channels every day with Shazam only. A library page is marked as such and shows how much of the set was identified. Anyone can run a full scan on it; when the full scan names more tracks, it replaces the library page. Library pages are only indexed once they hold a real tracklist.

Accuracy

On sets we measure against reference tracklists, a full scan identifies most tracks correctly and marks the rest as unknown rather than guessing. Fingerprinting has known blind spots: unreleased music, heavily pitched or layered blends, and records that are not in the engines' catalogues. Corrections from the set's owner and from the community feed back into future scans.

Corrections

Every row can be corrected by the set's owner. Corrections are kept with their source and can be reverted. They also improve later scans: a record you name once is recognised by its audio the next time it plays, in anyone's set.