Workflow: Batch QC & automation
File › Check a Folder… checks every audio file in a folder against Deliver to, with no script: a row per file with its verdict and readings, a report per file and a CSV of the rows. For a pipeline, the specula command-line tool, eight Shortcuts actions and three Finder services run the same measurement engine, so a script walking a directory gets the window's numbers as sorted-key JSON that diffs cleanly across runs.

Opening, reading and closing files by hand is fine for one master and slow for a folder of forty.
The slow part of the manual loop is not the measurement but the file handling around it: open, wait, read off six numbers, note the failures, next file. Specula runs that loop itself, in the app or from a script, and every path runs one engine, so a batch pass and a hand check of the same file return the same integrated loudness, true peak and dialogue loudness.
analyze, compare, edit, generate and report, for shell scripts that walk a folder of files.Choose the folder, check what it is judged against, run it.
Install once, then put the same numbers in a loop.
specula into /usr/local/bin (one admin prompt if that folder needs it), and specula --version confirms it. App updates keep it current. It shares the app's license and trial: it runs through the 7-day trial and needs the license activated in the app after that.specula analyze <FILE> emits the headline numbers as JSON: integrated LUFS, true peak, sample peak, loudness range, dialogue loudness and stereo correlation, with speech.map saying whether speech was detected or came from corrected regions saved for the file's length. --no-vad skips speech detection (faster), --no-stereo the correlation pass, and --no-sidecar measures with detected speech even where corrected regions are saved. Walk the folder, parse one field, flag the misses.specula report <FILE> runs the full report through the headless pipeline. --format json|html|pdf|csv picks the shape (PDF is the report the app's preview renders and needs --out; CSV is a flat metric,value table). --mode music|podcast|vod|broadcast picks the mode whose default targets judge the file: Spotify, Apple Music, YouTube and Tidal; Apple Podcasts and Spotify (Podcast); Netflix, Prime Video, Apple TV+ and Max; EBU R128 and ATSC A/85. It does not read the app's Deliver to or custom targets and has no audiobook mode, so ACX is judged in the app and in Check a Folder. --vad-threshold, --vad-min-speech and --vad-min-silence take the app's speech settings, and --no-curve drops the per-100 ms loudness curve for smaller batch output.specula compare <A> <B> reports every metric side by side with B − A deltas. --match-loudness nets out the integrated-LUFS delta so the deltas show spectral or shape differences instead of level drift; --out-diff <PATH> writes the A − gainB·B residual as a Float32 WAV and adds a residual block to the JSON. B at another sample rate is resampled to A's; the channel counts must match.specula generate --out <OUT> synthesizes a test signal and writes it: one source per run (--sine, --square, --saw, --triangle, --noise white|pink|brown, --sweep START:END, --impulse, --silence) across N identical channels, with --sample-rate, --bit-depth 16|24|32f and --container wav|aiff|caf. For layered or time-sequenced multichannel signals, a JSON spec goes in with --spec-file; the receipt on stdout reports the written format and the measured peak, RMS and integrated LUFS. The sources are the Signal Generator window's.A report straight to PDF, the same shape the app prints:
specula report master.wav --mode music --format pdf --out master.pdf
Every subcommand emits pretty-printed JSON with sorted keys, so runs diff cleanly: a changed number is the only thing that shows up. A stretch of digital silence reads "-inf". The app's codec and mono checks are not in the command line or the Shortcuts actions. Full command reference in the command-line section of the user guide.
Eight App Intents let the Shortcuts app and Siri call the same engine the command line does, with output a later step can chain on. Four actions ship in two variants each, one taking a file and one a path: Get Measurements, Compare Files, Get Compare Diff and Get Report.
The results are Shortcuts output to chain on. Get Measurements exposes every numeric field as a variable, so a later step can branch ("if Integrated LUFS > -12 then…") without parsing JSON. Compare Files returns both sides and the B − A deltas; Get Compare Diff returns the residual WAV, with B resampled to A's rate where they differ; Get Report returns the JSON, HTML or PDF file for a step such as Save File or Mail, judged in the loudness mode chosen in the step, as on the command line. The actions read corrected speech regions saved for a file, as the app does.
The (from Path) variants take the audio file as text, so a clipboard-driven shortcut is Get Clipboard → Get Report (from Path). They accept bare paths, paths with spaces, tildes, shell quotes, curly quotes from Notes or Mail, and percent-encoded file:// URLs.
Spotlight phrases trigger them too: "Measure with Specula", "Compare files in Specula", "Get diff with Specula", "Build a report with Specula", and their path-input siblings. The six measurement, compare and report actions start with speech detection off, a switch in the step turns it on for the dialogue loudness, and the two Get Compare Diff actions return a pure residual with no speech step.
For a right-click pass, three Finder services install with the app: Analyze in Specula, Compare in Specula and Get Report from Specula. The report service runs without a window: it measures the file, builds the PDF and puts it next to the source (or in ~/Downloads if the source folder is not writable), then reveals it in Finder.
One gotcha on the command line: negative numbers need the --option=value form. A bare leading - reads as an option name, so --gain -3 is rejected and --gain=-3 works (likewise --normalize-lufs=-14, --lufs-ceiling=-1). Positive numbers work either way.
The window, Check a Folder, the command line, Shortcuts and the Finder services run one measurement engine. The command line also fixes what a level move can: specula edit in.wav --normalize-lufs=-16 --lufs-ceiling=-1 --out out.wav normalizes and caps a file unattended, so a Folder Action can correct what it flagged. A noisy capture or a bad mix is left as it is, with numbers that diff across runs. The editing workflow