Audio loudness measurement receipt
3 deterministic fixtures × 3 reference LUFS levels = 9 cells. Measured via FFmpeg ebur128 filter.
Reference level explanation
- -14 LUFS - Spotify for Artists reference for music.
- -16 LUFS - Apple Podcasts reference for podcasts.
- -23 LUFS - EBU R 128 broadcast reference.
These are reference levels used to MEASURE loudness. They are NOT prescriptions for any platform. Per-platform loudness targets are a separate dataset (/data/per-platform-loudness-targets) sourced from each platform's published creator docs.
Run summary
9/9 cells PASS. Methodology: FFmpeg loudnorm filter (single-pass measurement).
Input LUFS per (fixture × reference level)
| Fixture | ref -14 LUFS | ref -16 LUFS | ref -23 LUFS |
|---|---|---|---|
| audio-sine-1khz-5s sine-1khz | -21.05 LUFS tp: -18.06 dBTP | -21.05 LUFS tp: -18.06 dBTP | -21.05 LUFS tp: -18.06 dBTP |
| audio-pink-5s pink-noise | -20.72 LUFS tp: -7.28 dBTP | -20.72 LUFS tp: -7.28 dBTP | -20.72 LUFS tp: -7.28 dBTP |
| audio-speech-5s speech-like | -9.05 LUFS tp: -4.47 dBTP | -9.05 LUFS tp: -4.47 dBTP | -9.05 LUFS tp: -4.47 dBTP |
Methodology
3 deterministic audio fixtures (sine 1 kHz, pink noise, speech-like composite) generated via FFmpeg lavfi at 48 kHz mono. Each fixture passed through FFmpeg loudnorm=I=-X:TP=-1.5:LRA=11:print_format=jsonwith the reference LUFS as I; the first-pass measurement returns input_i (measured LUFS-I),input_tp (true-peak dBTP). Reproducible via the documented script.
Reproduce: ./node_modules/.bin/tsx scripts/koldflux-task08/run-benchmark.ts
Honesty: This receipt documents how loudness is measured. It does NOT prescribe loudness targets for any specific platform.