agentleFS
MyDum-bsuMyDum-bsu/music-restoration/skills/audio-measure/SKILL.md

audio-measure

Skill4 starsChanged 11 days agoMITView on GitHub

Measure audio quality and authenticity — DR (TT convention), LUFS, true peak, clipping, bit depth/sample rate, fake hi-res and lossy-transcoded-to-FLAC. Use when asked to "measure DR", "what's the quality of this file/album", "is this real lossless?", "is it an upconvert?", "compare versions by quality", "check loudness/peaks", or as a step before picking a release version and before restoration. Do NOT use for finding or downloading releases (that's tracker) or for running the restoration itself (release-restore).

The file

Files
SKILL.md

audio-measure: audio quality and authenticity

One script, one answer to the question "what is this file, really".

python3 scripts/audio_measure.py <file|dir> [...]
python3 scripts/audio_measure.py --json out.json <dir>
python3 scripts/audio_measure.py --quick <dir>   # no spectrum/LUFS

Prints a table: format (sample rate/bit depth), DR, crest, LUFS, TP, spectral cutoff and a verdict. --json adds machine-readable output (all fields, including hf_floor_db).

What the fields mean

fieldmeaning
drDR in the TT convention (peak − RMS per blocks), the same metric as in the restoration corpus. Do not confuse with foobar DR: the numbers are close but not identical
crest_ttpeak − RMS over the whole file
lufs_iintegrated loudness (EBU R128) — "how loud"
tp_dbfstrue peak: above 0 dBFS means intersample peaks, common for crushed masterings
cutoff_hzthe last frequency where the spectrum is still within 55 dB of the peak
hf_floor_dbmedian of the spectrum above the cutoff relative to the 1–10 kHz band. This is the "wall" detector
clipped_pctshare of samples at full scale
fake_hiresclaims >44.1 kHz, but there is no content above 22 kHz
lossy_suspectcutoff below 19 kHz and floor above the cutoff < −75 dB (a real "wall")

How to read it (and what not to do)

  • lossy_suspect is a flag to verify, not a verdict. A dark or heavily limited master also gives a low cutoff, but no "wall" (the floor above the cutoff stays at −30…−40 dB). Even a sharp wall can originate in filtering or mastering; the floor does not rule out false positives. Verify acquisition provenance and the release edition. A control transcode demonstrates one detectable case, not general authenticity detection.
  • DR ≤ 6 — "heavily crushed", 7–9 — dense, ≥11 — healthy. This is not a "bad/good" judgement. Use it as one clue alongside specific audible defects.
  • TP > 0 dBFS does not mean the file clips: these are intersample peaks, potentially relevant to playback headroom. TP alone cannot establish audible harshness. The conservative runner uses a −1 dBFS ceiling on processed outputs.
  • --quick skips the spectrum — for large libraries; DR and clipping are always computed.
  • DR and spectrum use only the first 90 seconds. This can miss later defects and is not full-track acceptance. restore_safe performs full decode/QC.
  • fake_hires is a bandwidth suspicion flag, not proof of an upsample. Low ultrasonic content alone cannot establish the history of a recording.

Pitfalls

  • Bit depth is taken from ffprobe's bps, not from bits — otherwise the output shows ?.
  • LUFS/TP are parsed from the LAST summary block of ffmpeg: Integrated loudness and True peak appear in different places in the output, and Peak: occurs several times.
  • Comparing versions by DR only makes sense on the same material: different masterings of one album give different DR at the same loudness.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.