Where the numbers come from

AcousticBrainz — the key, the tempo, the loudness, the dynamics and the tuning. An open database of automatic audio analyses contributed by the public, released under CC0 1.0, the public-domain dedication. The file used here is the "lowlevel features" dump of 2022-06-23: 29,460,584 analyses of 7,564,449 recordings. The dump directory.

MusicBrainz core data — the names of the recordings, the artists and the releases, and the map of which recording ids are the same thing. Core entities are CC0 1.0; the canonical dump of 2026-09-17 is the one read. The dump directory.

What is deliberately not used: MusicBrainz tags, genres and ratings. Those are contributed under CC BY-NC-SA 3.0 — non-commercial, share-alike — which is not a licence an advertising-funded site may use. That is why there is no genre anywhere on this site and never will be from that source, and why there are no album covers or label artwork either.

Every file is pinned by SHA-256 in scripts/build-songs.mts, and the build refuses a file whose hash has moved. The pins:

• tonal — fa1d9e4c5e1af80372d9dbf96bc25bb84a3b0165600fbe2684496dbd1445bc0d
• rhythm — 5f813ff49c2ac1f35cca3947e081178758cddbd9b5292163dff58bdf3a025fad
• lowlevel — cf1b6511d9c33996edb1bbf1c6415b428bf45f263b0f87af4a7afa8ed70985ea
• canonical — 04a30bd748013851145c352f5a4c73f1bc3742a15a87eda413a50abd7610e1ed

AcousticBrainz counts no plays. What it has instead is how many people independently ran the extractor over a recording, and that is a real signal — an analysis is somebody who had the file. A recording is on this site when at least 125 analyses of it exist; 9,025 of the 7,564,449 recordings clear that bar, and the 7,600 with the most have pages here. 198 more were dropped because MusicBrainz says they are the same recording as one already listed.

That ranking is not a chart. It reflects the music AcousticBrainz's contributors owned between 2007 and 2022, which leans towards rock and towards albums people ripped — so the list is heavier on the 1970s to the 2000s than any listening figure would be. It is stated here rather than hidden because a list whose rule is invisible is an editorial pick pretending not to be.

A recording is filed under a key only when more than 50% of its analyses agreed on it. Its tempo page is the whole number its median reading rounds to. Both rules are printed under the lists they produce.

Measured up to 23.06.2022 — AcousticBrainz stopped collecting in 2022, so nothing released after that date is here. AcousticBrainz stopped accepting submissions in 2022 and the dumps have not moved since; the site cannot grow with new releases and does not pretend to.

Key. Essentia's key estimate, per analysis. The page shows the key the largest number of a recording's analyses agreed on, and the share that agreed. Automatic key detection confuses a key with its relative more than with anything else, so a low agreement usually means exactly that pair.

Tempo. Essentia's BPM estimate, per analysis; the page shows the median and the middle eight tenths. A detector that halves or doubles a tempo is common, which is why the median is published rather than the mean.

Time signature. Beats per bar are not in this dataset. AcousticBrainz's analysis reports a tempo, a key and a loudness, and no time signature at all, so this page does not print one — a 4/4 here would be a guess dressed as a measurement.

Loudness. Essentia's average_loudness: a number between 0 and 1, with no unit. It is not LUFS, it is not dBFS, and comparing it with a streaming service's loudness figure is meaningless.

Dynamics. Essentia's dynamic_complexity, in decibels: the average distance of the signal from its own global loudness. Low means a flat, heavily compressed master; high means a recording whose level still moves.

Tuning. The reference pitch the analysis inferred, in hertz, with its distance from concert A = 440 in cents. It describes the transfer — the pressing, the tape speed, the mastering — as much as the performance.

Analyses. How many people submitted an analysis of this exact recording to AcousticBrainz. It is the only popularity signal a CC0 audio dataset carries, and it measures what contributors owned, not what the world played.