Specula app icon

Version 2.0.1

Specula

Audio analysis and repair for macOS. Specula judges a file against where it is going: Spotify, Apple Podcasts, ACX, Netflix, EBU R128 or a spec of your own, with the verdict in a sentence and the fix that would change it. Seven tasks cover the rest of the work: inspect the file, edit it, compare up to six versions down to the residual, check audiobook chapters, correct speech regions, repair in the spectrogram with a squeak's overtones included, and record.

7-day trial, one-time purchase, free update for every 1.x license, macOS 14 and later

Specula's window in the Inspect task on a song: the Spotify delivery card with its verdict and fixes in the sidebar, the waveform, the loudness view with Spotify's target as a dashed line, and the spectrogram. The squeak under the interview selected on its lowest line, with its overtones outlined above it by Include harmonics, before Apply. The same view after Erase, with the squeak and all its overtones gone and the voice around them unchanged. The Chapters task on an audiobook: the chapter map over the chapter table with its ACX, Consistency and Details columns, one chapter failing in red, and the ACX summary and Level chapters in the sidebar. The same audiobook after Level chapters, with every chapter's ACX cells green. The Compare task with four versions of a song: the metrics matrix with each slot's readings and its difference from A, the waveform under it, and the slot in front's offset, level match and trim in the sidebar. The residual of slot B playing in the Compare task, with the residual's own peak and RMS in the sidebar. The Speech task on a podcast episode: detected and corrected speech regions on the waveform, the region tools in the task bar, and Level dialogue with its prediction in the sidebar. The Check a Folder window after a run: the folder, Judged against with its targets, the report formats, and a row per file with its verdict, integrated loudness, true peak, LRA and Speech %. A song after the music analysis: the Music group in the sidebar with the key and tempo, the section band with each section's loudness under the ruler in bars, and the Stems view with the vocal, drums, bass and other lanes.

The delivery check

Open a file and name where it is going. The delivery card says what that platform will do to it, in a sentence, with the integrated loudness and true peak under it and the fix that would change the verdict as a button.

A squeak under the interview

In the Spectral task a brush stroke along the squeak's lowest line selects it. Include harmonics outlines its overtones from 2× up in parallel, so the whole sound is one selection.

After Erase

Erase takes the selection out and leaves the voice around it as it was. The spectrogram shows the gap where the squeak was and nothing else different.

An audiobook, chapter by chapter

The Chapters task reads the chapter marks and judges each chapter against ACX: RMS, peak, noise floor and consistency across the set. One chapter fails, and the table says which cell.

Level chapters

Level chapters brings every chapter into the ACX window and into line with the others. The same table, every cell green, before the files go out.

Four versions of a song

Compare holds up to six versions, level-matched and aligned to the sample. The matrix shows each slot's readings and its difference from A, and one key switches which one plays.

Down to the residual

Play the residual and you hear only what differs between two versions, with the residual's own peak and RMS in the sidebar. If the two are identical, the residual is silent.

Speech regions

The Speech task finds where people talk, lets you correct the regions, and measures dialogue on its own. Level dialogue predicts the result before it applies.

A folder at a time

Check a Folder runs the delivery check over every file in a folder and writes one row per file, with a report per file and a CSV of the rows.

A song's sections and stems

On macOS 27 the music analysis reads a song's key, tempo and sections, and splits it into vocal, drums, bass and other. Each section's loudness sits under the ruler in bars, and the vocal is measured against the bed.

BS.1770-4, 48 kHz, 30 s generated program
0 −12 −24 −36 verse quiet chorus verse fade 0:00 0:10 0:20 0:30 target −14.0

Integrated

−14.7 LUFS

−0.7 LU vs −14.0

Short-term 3 s

−21.5 LUFS

Momentary 400 ms

−53.6 LUFS

True peak

−1.2 dBTP

sample peak −2.2 dBFS

What to watch. The quiet passage drops short-term by 10 LU and integrated barely moves: blocks more than 10 LU under the running mean fall out of BS.1770's relative gate. True peak ends a decibel above sample peak - the inter-sample overshoot that only shows up once the 4× oversampler looks between samples. Playback runs at 2×.

22

delivery targets in five categories, from streaming to broadcast

6

versions compared at once, level-matched and sample-aligned

4×

oversampled true peak, judged against each target's ceiling

−60 dB

ACX's noise floor, checked on the room before the take

Deliver to

Every file is judged against where it is going, in a sentence, with the fix that would change it.

Monitors won't flag what delivery QC will: integrated loudness half a dB over Spotify's target, a dialogue mix four LU under Netflix's speech-gated level, one channel with its polarity flipped that nobody caught on headphones. Those numbers are already in the file.

Deliver to names where the file is going, Music streaming, Podcast, Audiobook, Film and TV or Broadcast, and the targets checked there, one of them primary. The delivery card under it gives the primary target's verdict as a sentence, such as "Spotify turns this down by 5.0 dB." or "Apple Podcasts plays this as recorded, 8.7 dB under its −16 LKFS reference.", with the reading in the target's own terms: integrated loudness, dialogue loudness, or ACX's RMS and noise floor. The card judges the whole file as soon as it has loaded.

Fixes in words. The primary target's fixes are buttons that say what they do, such as Limit peaks to −1.0 dBTP, Normalize to −14 LUFS and Hear at Spotify's level, offered only where they change something. Every other checked target keeps its fixes in a Fix menu.

Two checks for streaming. Under Music streaming the card also says whether a stereo master folds to mono, such as "Mono-compatible. Correlation +0.62 over the file.", and what Apple Music's AAC encode does to it: the true peak and the samples over full scale after the encode, with Limit peaks set to leave room for it.

ACX in its own words. A volume between −23 and −18 dB RMS, peaks less than −3 dB and a noise floor less than −60 dB RMS: a reading exactly on a limit fails, as ACX states it, and Limit peaks for ACX aims at −3.1 dBTP.

Music streamingSpotify, Apple Music, YouTube, Tidal, Amazon Music, Deezer, SoundCloud, AES TD1008
PodcastApple Podcasts, Spotify (Podcast), ACX
AudiobookACX leading, with Apple Podcasts and Spotify (Podcast)
Film and TVNetflix, Prime Video, Apple TV+, Disney+, Max, dialogue-gated at −27 LKFS
BroadcastEBU R128, R128 live, R128 S1 short-form, ATSC A/85, ARIB TR-B32, OP-59

A spec of your own joins its category as a custom target: a reference level, a tolerance, a true-peak ceiling, and integrated, dialogue-gated or RMS measurement. Every target's reference, tolerance and ceiling is listed in the user guide.

The delivery card for three sample files

Deliver to Spotify · Music streaming

Spotify

-14 LUFS · ≤ -1 dBTP

Spotify turns this down by 5.3 dB. True peak 0.4 dBTP is over its -1 dBTP ceiling.

Integrated-8.7LUFS

True peak0.4dBTP

Limit peaks to -1.0 dBTPNormalize to -14 LUFSHear at Spotify's level

Also checked

  • Apple Music-7.3 dBFix
  • YouTube-5.3 dBFix
  • Tidal-5.3 dBFix

The readings are typed in for the demo. The sentences and the fixes are the app's, made by its own code.

The Deliver to picker: the five delivery categories on the left, the chosen category's targets on the right with a checkbox each and the primary target marked.
Deliver to. Five categories, their targets, and the one the verdict leads with.
The delivery card for a loud master: the verdict as a sentence, the integrated loudness and true peak, and the Limit peaks and Normalize buttons.
The delivery card. The verdict as a sentence, the readings in the target's terms, the fixes as buttons.

Seven tasks

One task at a time, each with its own tools, its own views and its own part of the sidebar.

The task switcher sits in the top bar, ⌘1 to ⌘7 reach the tasks, and Esc, with nothing left to clear, returns to Inspect. The file, its zoom and its scroll stay as they were from task to task.

Inspect

⌘1

The file as it is. The waveform, the loudness with the primary target drawn as a dashed line, and the spectrogram computed for the whole file. The sidebar holds the file, the delivery card, Loudness details with a Live side for what is playing, and the speech, music and channel readings. ⌘Return measures a selection into a card of its own.

Edit

⌘2

Cut, trim, fades and gain on the task bar; Normalize, Limit True Peak and Level Dialogue in its Loudness menu; a chain of Audio Unit effects rendered as one edit. A selection shows its length and its common edits right over the waveform. The sidebar lists the history: every edit by name, up to 16 steps back, any of them a click away. Save As writes a new file, and saving over the original is a setting.

The Edit task on a song: a selection with its length and edits over the waveform, the Edit task bar, and the sidebar's Edit part listing the history of four edits.
Edit. The task bar, a selection's edits over the waveform, and the history in the sidebar.

Compare

⌘3

Up to six versions of a mix or a master sit in a matrix of their readings, each with its difference from A. Level match, sample-accurate Align, and the residual: everything that differs between A and a slot, playable with D and measured on its own. A file at another sample rate switches in without a pause, and Apple Music's encode of the master can take a slot of its own.

The Compare task with four versions of a song: the metrics matrix with each slot's readings and its difference from A, the waveform below, and the slot card for slot B in the sidebar.
Compare. Four versions in the matrix, level-matched, with the slot in front's alignment in the sidebar.

Chapters

⌘4

Chapters found at the silences, then a table of every chapter's readings with ACX's marks, an ACX summary for the book, and Level chapters, which brings every chapter to a common RMS or LUFS and shows each chapter's gain before it is applied. Chapters can also come from a song's sections.

The Chapters task on an audiobook file: the chapter map over the chapter table with ACX's marks, and the ACX summary and Level chapters in the sidebar.
Chapters. Every chapter against ACX, and the gains Level chapters would apply.

Speech

⌘5

Silero's speech regions, corrected by hand: add, split, mark in and out, delete. The dialogue loudness and the speech percentage follow the corrected regions. Level dialogue predicts the dialogue loudness and the noise floor before it is applied, and judges the floor against ACX.

The Speech task on a narration: corrected speech regions over the waveform, the Speech task bar, and Level dialogue's prediction in the sidebar.
Speech. Regions corrected by hand, and Level dialogue's prediction before Apply.

Spectral

⌘6

The spectrogram at full resolution, with a rectangle, lasso, brush or wand to select a cough, a squeak or a hum. Include harmonics adds the selection's tone at 2×, 3× and up, where it rings, so one stroke on a squeak's lowest line takes its overtones too, following the squeak as it glides. Attenuate it, erase it to the noise floor around it, or repair it from the audio either side; Solo plays the selection and Hear Result plays the edit before Apply. Audio beyond the selection and its feather stays bit for bit.

The Spectral task on an interview with a chair squeak: a brush stroke along the squeak's lowest line, its overtones from 2× to 7× outlined above it, the tool column on the left edge, and the Spectral edit panel in the sidebar with Include harmonics ticked.
Spectral, with Include harmonics. One stroke on a squeak's lowest line, and its overtones from 2× to 7× outlined above it, ready for Erase.

Record

⌘7

Any input, armed with live meters and a room check against ACX's noise floor before the take. A take lands as a new file, appended, inserted at the playhead or punched in; a punch at the playhead keeps the file's length, and speech is detected again where it landed. Takes stay for seven days.

The Record task armed: the input's level, the momentary loudness and true peak, the room check against ACX's noise floor, and the takes in the sidebar.
Record. The input armed, with the room checked against ACX before the take.

Live meter

The standard itself, running in this page.

The meter at the top of this page is a working BS.1770-4 meter. The page synthesizes a 30-second program - verse, quiet passage, loud chorus, verse, fade - and meters it as it plays: K-weighted momentary and short-term, gated integrated loudness, true peak at 4× oversampling. Specula runs the same standard natively, on your files.

The pipeline

master.wav mono to 7.1 BS.1770-4 · M / S / I / LRA true peak · 4× oversampled FFT spectrum · spectrogram Silero VAD · speech-gated LUFS correlation · phase scope 22 delivery targets verdicts in words Reports · CLI JSON, HTML, PDF, CSV master.wav mono to 7.1 BS.1770-4 · M / S / I / LRA true peak · 4× oversampled FFT spectrum · spectrogram Silero VAD · speech-gated LUFS correlation · phase scope 22 delivery targets verdicts in words Reports · CLI JSON, HTML, PDF, CSV
The analysis pass. A loaded file fans out into the five paths; verdicts and reports read from that same pass, in the window or from a script.

A folder at a time

Check a Folder (File › Check a Folder…) checks every audio file in a folder against Deliver to: one row per file with its verdict, integrated loudness, true peak, loudness range and Speech %, measured with the app's speech settings and any saved speech regions made for the file. Reports in JSON, HTML or PDF are written for each file as it finishes, and a CSV of the rows at the end, into a new folder beside the checked one and never over a file.

The same engine runs from the terminal: five subcommands, analyze, compare, edit, generate and report, return the numbers the app shows as deterministic sorted-key JSON. Install it from Specula › Install Command-Line Tool…

specula
$ specula analyze master.wav
$ specula compare revA.wav revB.wav --match-loudness --out-diff residual.wav
$ specula report mix.wav --mode music --format pdf --out report.pdf

Shortcuts and the Finder's Services menu reach the same engine. The automation workflow covers all three; the full flag reference is in the user guide.

The Check a Folder window after a run over a folder of masters and speech files: one row per file with its verdict, integrated loudness, true peak, loudness range and Speech %.
Check a Folder. Every file against Deliver to, with a report per file and a CSV of the rows.

Songs

On macOS 27, Apple's on-device music analysis reads a song's key and tempo, its sections with each one's loudness under the ruler, its bars and beats for the ruler and for snapping, and its four stems: vocal, drums, bass and other. A detected song opens the sidebar's Music group and switches the ruler to bars; the Stems view shows where each stem plays, and stem-gated loudness measures the vocal against the bed. A song's sections can become chapters for a per-section loudness table. The analysis is labelled as Apple's throughout, and the loudness figures stay Specula's own BS.1770-4 measurement.

A song after the music analysis: the Music group in the sidebar with the key and tempo, the section band with each section's loudness under the ruler in bars, and the Stems view with the vocal, drums, bass and other lanes.
A song. Key and tempo, sections with their loudness, the bars ruler and the stems.

Built for your job

Eleven ways into the same engine, each with the tasks it uses.

Spoken-word work is where Specula goes deepest

Neural speech detection you can correct by hand, dialogue-gated targets for every major platform, every audiobook chapter against ACX, and the room checked before the take.

Everything it measures

Every file is measured whole as it opens; Live readings follow playback; any selection on request.

EBU R128

Momentary, short-term, integrated, LRA and true peak, BS.1770-4 to the EBU test set.

Delivery targets

22 targets in five categories, plus your own; verdicts and fixes in words.

Speech-gated loudness

Silero neural speech detection, corrected by hand, gating the dialogue reading.

Waveform

Every channel, amplitude fit, clip detection, speech regions overlaid.

Spectrum and phase

Five window types, note names, band filtering; the whole file's average at rest.

Spectrogram

The whole file, four colormaps, log or mel, Floor and Ceiling knobs.

Stereo and mono

Left/right correlation over time, and a mono check for streaming masters.

Codec check

Apple Music's AAC encode of the master: its true peak and samples over full scale.

Per-channel statistics

Peak, RMS, crest, DC offset and clipped samples per channel.

Loudness thresholds

M, ST and TP limits that flag the whole file's curves.

Selection analysis

⌘Return measures any selection into its own card.

Check a Folder

A folder against Deliver to, with a report per file and a CSV.

Plugin chain

Audio Unit effects in Edit, rendered as one undoable edit.

Spectral editing

Select in time and frequency, harmonics included; attenuate, erase or repair.

Recording

New file, append, insert or punch, with a room check.

Signal generator

Tones, noise, impulses and sweeps, stacked per channel.

Channel routing

N to M output matrix, mixdown presets, render to file.

Music analysis

Key, tempo, sections, bars and stems on macOS 27.

Reports

PDF, HTML, JSON and CSV from one export sheet.

Formats

WAV, AIFF, FLAC, CAF, MP3, AAC and raw PCM; mono to 7.1 and beyond.

Command line

analyze, compare, edit, generate and report, as sorted-key JSON.

Shortcuts and Finder

Eight App Intents and three Finder services.

The whole app, task by task

Every task, view and setting, what each reading means, the delivery targets' levels, and the complete keyboard map.

Open user guide

Standards

Specula implements the actual specifications, not approximations of them.

EBU R 128 loudness logo EBU R 128
loudness compliant
ITU-R BS.1770-4K-weighting filter (two cascaded biquad IIR sections), 100 ms gating blocks, dual gating (−70 LUFS absolute, −10 LU relative), 4× oversampled inter-sample peak detection.
EBU R128Momentary (400 ms window), Short-term (3 s window), Integrated (gated whole-programme), LRA (10-95 percentile range), True Peak (dBTP). Loudness validated against the EBU R128 loudness test set (Tech 3341/3342).
Silero VADNeural voice activity detection (MIT). Runs at 16 kHz on every loaded file, classifies speech/non-speech per 100 ms block, and feeds the speech-gated loudness path.
FluidAudioApache 2.0 Swift wrapper for Silero on Apple platforms. The model ships inside Specula, so speech-gated analysis runs fully offline with no download.

Requirements

macOS
Sonoma 14.0 or later. Music analysis needs macOS 27; the HDR accents need macOS 26 and a display with HDR headroom.
Hardware
Any Mac that runs macOS 14. Apple Silicon recommended for large files; uses Apple's vDSP throughout.
ⓘ

All audio is processed locally. Specula never transmits your audio anywhere.

FAQ

Is Specula 2.0 a free update?

Yes, for every Specula license. Specula updates itself through Check for Updates… in the app menu, and 2.0 keeps the settings 1.x had.

Where did the Mode, Dialogue mode and Chapter mode go?

Deliver to replaces the Mode, with five categories in place of four. Dialogue mode is the Speech task (⌘5) and Chapter mode the Chapters task (⌘4). The user guide's Coming from 1.4 section lists where everything went, shortcuts included.

Do I have to analyse a selection to get the verdict?

No. The delivery card judges the whole file as soon as it has loaded, and the reports read the same pass. ⌘Return measures a selection into a card of its own when a part of the file is what you need.

Can Specula check a whole folder?

Yes. File › Check a Folder… checks every audio file in a folder against Deliver to and writes a report per file and a CSV of the rows. The specula command-line tool and the Shortcuts actions run the same engine from a script. See the automation workflow.

Does my audio leave the Mac?

No. Loudness, speech detection and the music analysis run on the Mac. The app's only outbound connections are update checks and license validation, and neither carries audio. See the privacy policy.

More questions

Before delivery

Specula is the last check between the mix and the delivery. It measures what a listening pass can miss: a phase problem, a true peak half a dB over its ceiling, a dialogue mix three LU off Netflix's target.

The user guide covers loudness gating from the start, and its keyboard reference lists every shortcut.