Workflow: Compliance
A delivery spec is a hard line, and a file either clears it or comes back. Under Film and TV or Broadcast, the delivery card judges the whole file against every checked spec as soon as it has loaded, and says why one fails in a sentence, with the reading and the limit. The thresholds under the loudness graph mark where, and the PDF, HTML or JSON report states what the file was judged against.

A delivery spec is a loudness window, a true-peak ceiling, and sometimes a short-term cap or a noise-floor limit.
Miss any one and the file bounces. EBU R128 S1 caps max short-term. Netflix measures loudness over dialogue, not the whole programme. A true peak that reads fine on the samples can go over once the peaks between them are counted. When a file opens, Specula measures the full-programme integrated loudness and the dialogue-gated loudness, the 4× oversampled true peak, max short-term and the noise floor in one pass, checks each against every checked spec, and names the criterion that fails and by how much.
Deliver to the category, read the card, find the breaches, fix, export.
Not compliant · S↑ (−15.0 LUFS)) and say what speech was measured with. The JSON carries a delivery block and a verdict per target in loudnessTargets.targets[] (status, actualLUFS, the reason and the limit), so a delivery pipeline can branch on pass or fail. Right after an edit, Export waits until the file has been measured.A maximum says the master breached, and by how much. The marks say which second of forty minutes.
A QC note says "too loud". The marks say where: 2:14, 7:48, 31:02, and on which meter.
Max momentary, max short-term and max true peak settle whether the file ships, not which moment to touch. With the spec's ceilings set in the fields under the loudness graph, every 100 ms block that crosses one is shaded along the waveform, so each overshoot is a place to click to.
Three ceilings on three time scales, each catching a failure the others miss:
Each threshold shades in its own colour and layers over the others, so where two breaches overlap both show, and each field carries its count. The marks cover the whole file as soon as it has been measured. Above the waveform, the loudness graph plots momentary and short-term against the same thresholds, so each breach shows how far over it goes, not only that it crossed.
That is the distance between a QC report and a fix: the one shouted line to ride, the one transient to cap, while the rest of a nearly-right master stays as it is.


Film and TV and Broadcast are the hard pass or fail. Film and TV judges dialogue loudness in the −27 LKFS window that Netflix, Prime Video, Apple TV+, Disney+ and Max share, with a −2 dBTP ceiling. Broadcast judges EBU R128, its live variant and R128 S1 short-form (which adds the max short-term ceiling), ATSC A/85 (the CALM Act), ARIB TR-B32 and OP-59. Each spec carries its own reference, tolerance and true-peak ceiling; every number is in the user guide.
Music streaming deals in gains: each platform's verdict says what it does to the file (mastering). Podcast judges Apple Podcasts and Spotify (podcast), and Audiobook leads with ACX, judged in ACX's own words (audiobook).
Dialogue-gated specs read the speech-gated path. Netflix and its peers, ATSC A/85 and the podcast targets measure loudness over speech only, found by the Silero neural speech detector when the file opens, so room tone and music beds do not move the verdict. The picker lists them as "dialogue −27 ±2 LKFS", and the card's headline is the dialogue loudness. Where the detection is wrong, the Speech task (⌘5) corrects the regions by hand, and the file is measured again, so the card, the fixes and the reports follow the corrections.
A spec of your own joins its category as a custom target: a reference level, a tolerance, a true-peak ceiling, and an integrated, dialogue-gated or RMS measurement.
EBU R 128


A failing spec names its criterion and offers its fix where a gain or a limiter can make one; a short-term spike is marked on the timeline, and a noise floor reads against its limit. The passing file leaves with a report that states what it was judged against. The editing workflow