Audiobooks, scored chapter by chapter.

ACX checks three things and rejects on any of them. Two are easy to hit; the third is the one that catches people.

Loudness and true peak you can usually eyeball in a DAW. The room-tone noise floor is different: it's inaudible on headphones, invisible on a standard loudness meter, and a common reason an audiobook bounces back. Specula reads all three off the file directly, so the rejection condition is a number on screen instead of a surprise.

The workflow

Load to receipt in four steps. One file, one pass, every chapter scored.

A file-level pass isn't enough for a book. An audiobook can sit at a perfect −20 dB RMS overall while one chapter recorded on a different day drifts to −16, or a single quiet pickup dips under the noise-floor line. ACX checks per chapter, so Specula does too: every chapter gets its own ACX verdict on RMS, true peak, and noise floor.

The outlier flag catches consistency problems the absolute gates miss. Each chapter is compared to the book's own median loudness; anything beyond the tolerance (default ±2 LU) is flagged even if it's technically inside the ACX band. That's how you spot the chapter that will sound jarringly louder to a listener mid-book.

The corrected dialogue regions from step 1 feed straight into this: per-chapter speech-gated LUFS and the noise-floor reading all measure the speech and silence you confirmed, not what the raw VAD guessed.

Batch a whole series

Skip the app window when you're checking dozens of files.

The specula CLI runs the same engine the app uses, so a script gets the exact same numbers. Score one chapter to a PDF:

Or wire the Get Report App Intent into a Folder Action so every new chapter scores itself on save, and a downstream Shortcut step branches on Integrated LUFS or Speech-Gated LUFS without parsing JSON. Walk the folder, flag the chapters that miss, AirDrop the failures to the narrator. Full command reference in the CLI section of the user guide.

Each red cell points at its fix, and the two that bounce the most audiobooks have a fix in Edit mode.

Noise floor over −60 dB RMS is a common ACX rejection, and it's the one a uniform gain can't fix, because raising the voice raises the room tone with it. Level Dialogue (Edit mode) is built for exactly this: it ducks the room tone between phrases using the speech regions you confirmed in step 1, with a fast attack and slow release so word onsets and tails stay clean. Tick Lift dialogue to target and it also brings the voice to your RMS target in the same pass, capping inter-sample peaks at a true-peak ceiling. Because the silence is gated before the makeup gain goes on, the floor ends up lower, not higher, the move a plain Normalize can't make. It's a gate, not a denoiser: it rescues a marginal floor (a quiet room tone just under the gate), not a hissy recording, which still wants a better capture.

A chapter that drifts off the book's loudness is the consistency flag, and Level chapters (Chapter mode → Level…) is the one-move fix: pick the metric (RMS, the ACX one, by default) and the target (the book's own median, or a fixed value such as −20 dB RMS, mid the ACX −23..−18 band, for consistency and compliance at once), and Specula applies a per-chapter gain and re-measures. A true peak over −3 dBTP is a Limit TP pass that caps the inter-sample peaks without touching loudness. Each is one undoable edit that saves a new file, never the original. See the editing workflow →

Specula is the pass between "the narration is done" and "it's uploaded to ACX." It tells you, per chapter, exactly which files clear the spec and which won't, and gives you the fix for the two flags that bounce the most books: Level Dialogue for a marginal room-tone floor, and Level chapters for a chapter that drifts off the book's loudness. Read the numbers, fix the flags, hand off the receipt.