Specula

The standard itself, running in this page.

Below 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.

A song opens with its key and tempo, its sections and their loudness under the ruler, a bars ruler, beat snapping, and the vocal measured against the bed in LU.

On the timeline. Apple's on-device music analysis runs in the background after the file loads, about 0.6 seconds per minute of audio. Once it has found a song, the sidebar's File section reads the key and the tempo, a band under the waveform's ruler shows the song's sections with each section's integrated LUFS (red when a section sits outside the chapter deviation threshold from the song's median), the ruler's mode button gains a bars setting, the transport reads the bar and beat under the playhead, and a SNAP chip in the control strip lands selection edges on the nearest beat or bar. A click on a section selects it.

Stem-gated loudness. The sidebar's Loudness section gains a Stem-Gated block: integrated LUFS over the blocks where the vocal is present, over the bed without it, and their difference in LU, then the same three readings for drums, all Specula's own BS.1770-4 integrated measurement gated by Apple's stem presence spans. A side with less than 10 seconds of audio reads "-". A Sections block next to it lists the loudest and quietest sections and their spread.

Music module (⌃7). The detail view. Four stem lanes, Vocal, Drums, Bass, and Other, mark where each stem is present and how active it is over time. Above them sit the beat and bar grid, the song's sections, segments, and phrases as labelled bands and ticks, and Apple's pace figure per section. Click or drag on the module to place the playhead; scroll zooms the time axis with the other sections. Music analysis guide →

Section loudness map. Chapter mode takes its chapters from the song's sections, one per section tiling the whole file, and Analyse Loudness then reads integrated LUFS, true peak, RMS, LRA, and the deviation from the song's median for every section, in the chapter report and CSV as well.

Any file up to 15 minutes (a limit you can change) is analysed after it loads, in every Mode, and longer files are left alone until you ask; Settings → FFT / Analysis → Run music analysis chooses Songs, Every file, or Never, and View → Analyse Music runs it by hand. A key found means the file is a song: the Music module opens on its own, and if the Mode is not Music the sidebar offers a one-click switch. Nothing is uploaded. The JSON report carries the full result under musicAnalysis and the stem-gated readings under stemGatedLoudness, the PDF and HTML reports get a Music summary, and the CSV export gets Key and Tempo rows.

The analysis is Apple's Music Understanding framework and is labelled as Apple's analysis throughout; the loudness figures stay Specula's own BS.1770-4 measurement. It needs macOS 27. On earlier systems the timeline stays as it is and the module and its menu items are absent; everything else in Specula runs on macOS 14 and later.

Specula shows what's in the file, in numbers and in pictures, before it ships.

Monitors won't flag what delivery QC will: integrated LUFS half a dB over Spotify's target, a dialog mix four LU under Netflix's speech-gated number, a polarity flip in one channel that nobody caught on headphones. Those numbers are already in the file. Specula reads them out.

Comparison has the same blind spot. Was rev 4 better than rev 3, or just half a dB louder? Specula loads up to six versions, normalises them by integrated LUFS, optionally inverts polarity, and computes a sample-accurate residual: everything that differs between two takes, audible on its own.

Before a file leaves the studio, one window confirms the loudness, the true-peak ceiling, the platform compliance, the phase, and the spectral content of the master you're actually sending.

Everything it measures

Every measurement, on every file you load. Live during playback, offline on any selection.

The same engine ships as a CLI. 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 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

Batch QC, CI hooks, and the eight App Intents are covered in the automation workflow; the full flag reference is in the user guide.

Four delivery modes, twenty-two platforms and standards. The gain each will apply to your master, or a hard pass/fail against the spec.

Music mode shows the streaming penalty rather than a verdict. A master at −10 LUFS reads "Spotify −4 dB"; that's the exact gain Spotify will apply at playback to hit its −14 LUFS reference. The dot is green within ±1.5 dB of zero, yellow within ±4 dB, orange beyond, and a separate triangle lights when the true peak breaches the platform's ceiling.

Quiet masters get the honest readout. Apple Music's Sound Check, YouTube, and Tidal are down-only: a −18 LUFS master plays at −18, and the panel reads "as-is" in green instead of promising a boost that won't come. Spotify, Amazon Music, Deezer, and SoundCloud do boost quiet tracks, and there the +X dB number is real.

VOD and Broadcast modes are hard pass/fail: a green dot means loudness sits inside the tolerance band and true peak is at or below the ceiling. FAIL L marks a loudness band miss, FAIL TP a true-peak overshoot, FAIL S↑ a max-short-term overshoot (EBU R128 S1's tighter short-form spec), FAIL NF an ACX noise-floor breach. When several gates fail at once, the loudness miss is the one named first.

Dialog-gated targets (Netflix, Prime Video, Apple TV+, Disney+, Max; Apple Podcasts, Spotify Podcasts; ATSC A/85) evaluate against the speech-gated path and lead with dialog · in the panel. A file with no detectable speech is labelled "no speech" rather than scored against a gate it can't meet. Pick up to four targets per mode in Settings → Targets.

Match to target is non-destructive audition: press a target row's toggle and playback shifts to that target's level with a true-peak limiter at the spec ceiling. Press it on Netflix and hear the file the way Netflix will play it; release, and unity gain returns. Normalize commits instead: one uniform gain to land the target, a two-pass limiter to hold the ceiling, saved as a new file. When only the true peak is over, one-click Limit TP caps the inter-sample peaks without touching loudness.

For audiobooks, the common ACX rejection is room tone above the −60 dB RMS ceiling. Specula shows a live Noise Floor readout over every non-speech sample, and the ACX preset scores it as part of the verdict: above −60 dB, the target lights FAIL NF. Chapter mode runs the full ACX set per chapter; the audiobook workflow has the details.

The panel tracks playback live. For a broadcast delivery you'll usually want the offline analysis (⌘Return) so the verdict covers the entire master, not a rolling estimate. Every target's reference, tolerance, and true-peak ceiling is listed in the user guide.

Standards

Specula implements the actual specifications, not approximations of them.

EBU R 128
loudness compliant

Specula doesn't replace your DAW, your mastering chain, or your ears. It catches what they miss: the hidden phase issue, the half-dB true-peak overshoot, the dialog mix three LU off Netflix's target. Load the file, read the numbers, then send it.

Take the long way through the user guide if you're new to loudness gating, or skim the keyboard reference if you just want to drive it.