The small fix, without booting a DAW.

The fix is thirty seconds of work. Opening a DAW to do it isn't.

You loaded the file to check the numbers and now you can see the one thing that needs fixing: a few seconds of room tone to trim off the head, a fade to clean up the tail, a normalize to the LUFS target the delivery wants. Round-tripping that through a DAW means a new project, an import, a bounce, and a settings dialog for each. Doing it by hand on the master means there's nothing to fall back to if you get the gain wrong.

Specula does the small jobs where the file already is. Two rules keep it safe: edits are undoable (up to 16 levels of undo/redo), and saving always produces a new file. Specula has no in-place overwrite, so the source you loaded stays exactly as it was on disk.

Three steps, no project

Enter the mode, run the operation, save a new file.

A peak normalize that only watches sample values is watching the wrong ceiling. The number that gets a master rejected, and the one streaming platforms enforce, is the inter-sample peak: the level the signal reaches once the converter reconstructs the analog waveform between samples. A file can read −1.0 dBFS on sample magnitude and still overshoot when it's played back.

Specula's limiter measures the 4× oversampled inter-sample peak and enforces the ceiling against that, in dBTP, not against the raw sample magnitude. Normalize LUFS runs it automatically whenever the normalize raises loudness, in two offline passes: pass one builds the gain envelope from the input's oversampled peaks; pass two catches any residual peaks the release stage might leave fractionally over the ceiling.

Limit TP is the same limiter on its own, no loudness measurement, no gain stage. Reach for it when the integrated LUFS is already right and only the inter-sample peaks need capping under a platform ceiling.

Edit mode has keyboard shortcuts, active only in Edit mode so they don't shadow the transport keys, and listed in the Edit menu so the keys are always visible. One group acts on the selection, the other at the playhead.

Level Dialogue is the edit for spoken word. It's a region-aware downward expander keyed to the speech regions Specula already detects (the ones you can fine-tune in Dialogue mode): it ducks the room tone between phrases with a fast attack and slow release, so word onsets and tails stay clean, no clipped starts, no chopped ends.

Tick Lift dialogue to target and it also gains the speech to a loudness target in the same pass, capping inter-sample peaks at a true-peak ceiling. Because the silence is gated first and the makeup gain goes on after, the noise floor ends up lower, not higher, which is the move a plain dialog-gated Normalize can't make (a uniform gain lifts the floor along with the voice). That's why this is the fix for the most common ACX rejection, a room-tone floor sitting just over the line.

It's a gate, not a denoiser. It rescues a marginal floor, a quiet room tone just under the gate, not a genuinely hissy recording, which still wants a better capture or a dedicated denoiser. Like every edit here, it's undoable and saves as a new file. See the audiobook and podcast workflows for where it fits.

Edit mode hosts a chain of Audio Unit effects. Add any effect installed on your Mac (both AUv2 and AUv3), chain several in order, bypass or reorder them, and open each plugin's own window, or a generic parameter panel for plugins that ship no interface of their own. It works on mono and stereo files.

You tune by ear while the file plays, and every meter moves with you. Because the analysis tap sits after the chain, LUFS, true peak, FFT, spectrogram, and phase all read the processed sound as you turn knobs, so you're EQing or compressing against the same numbers you'll deliver against.

Render bakes the chain into the audio through fresh, latency-compensated instances and commits it as a normal edit you can undo. With the effect tail off (the default) the rendered audio is exactly as long as the input, so it still lines up for A/B and auto-align. Rendering clears the chain and remembers it, so Undo then Restore Last is the tweak-again loop, and it survives quitting the app. Render to Compare Slot loads the processed version next to the original in Compare mode so you can measure the difference with real numbers. AUv3 plugins run in their own process by default, so a crash in one of them can't take Specula down; AUv2 plugins run inside the app.

Spectral mode (⌘5) is for the blemish inside the audio you want to keep: a cough inside a piano chord, a chair squeak under speech, a bite of mains hum. Select it on a full resolution, zoomable spectrogram, in time and frequency, then attenuate, erase, or repair just that region, leaving everything else untouched. It's a bigger job than the trims and fades on this page, so it gets its own workflow: selection tools, the wand's dB tolerance and time reach, feathering, and audition before you commit.

Edit mode is for the jobs that don't earn a DAW session, most of them a single keystroke. Specula runs them on the loaded buffer, keeps 16 levels of undo/redo, and saves a new file every time, so the fix is fast and the master is never the thing you edited.