Specula app icon

Workflow: Reference comparison

Is rev 4 better, or just louder?

Half a dB of level reads as better to the ear. The Compare task holds up to six versions of a mix or a master, level-matched by integrated loudness and aligned to the sample, with every reading in one matrix beside its difference from A. The residual, A minus a version, plays and draws what the two do not share.

7-day trial, one-time purchase, macOS 14 and later

Specula's 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.
Four versions in the matrix. Every reading beside its difference from A, the waveform under it, and the slot in front's offset, level match and trim in the sidebar.

Why a raw A/B misleads

The louder version sounds better, so a straight switch measures the level, not the change.

Half a dB louder reads as better to the ear, every time, so a raw switch between rev 3 and rev 4 answers the wrong question. With the level out of the way, what is left is the change itself: the EQ move, the limiter, the new master chain. Compare matches the versions by integrated loudness first and lines them up to the sample, so the switch and the residual carry only that.

Set up a fair A/B

Open the versions, take level and timing out of the way, then listen to what is left.

1. Open up to six versions
Opened together, or dropped on the window together, files fill slots A to F in file-name order, Take 2 before Take 10, and the first is A, the reference. Add in the task bar, or File › Add Files to Compare… (⇧⌘O), adds more. 1 to 6 switch slots, and each column of the matrix has its own Listen button. A slot chip, dragged, reorders the slots, and its menu makes it the reference: the other slots' alignments follow the new A. A file at another sample rate, such as a 44.1 kHz stream beside a 48 kHz master, plays at A's rate, resampled once when it is added, so switching to it is instant.
2. Match the level
Match in the task bar (Compare › Level Match) plays every slot at A's integrated loudness, and the residual subtracts at the same gain. The slot card in the sidebar shows the gain it applies, and Trim adds a hand offset, typed or in steps of 0.1 and 1 dB, up to ±24 dB.
3. Align to the sample
Align (Compare › Auto-Align to A) cross-correlates the slot with A, first on a decimated signal to find the offset and then at the full rate to land on the sample, for takes that start within about 5 seconds of each other. The slot card's nudges move it by a sample or a millisecond, and its offset takes a typed value in samples. Invert polarity flips the slot.
4. Listen to the residual
Residual in the task bar, or D (Playback › Listen to the Residual), swaps the slot in front for A minus that slot: it plays and draws the residual, and the sidebar shows the residual's own peak and RMS. With the level matched, two versions of the same master net out to near silence, and what stays above it is what changed. A change to the slot's offset, gain, polarity or level match, or an Align, computes the residual again. Export Residual… (⌃⌘E) writes it as a WAV, and the export sheet can write it beside the reports.

What it looks like

The residual of slot B playing in the Compare task, with the residual's own peak and RMS in the sidebar.
The residual. A minus B, level-matched, with its own peak and RMS in the sidebar.
The Compare matrix: a column per slot with integrated loudness, true peak, sample peak, LRA, DR and max momentary and short-term, each with its difference from A, and a Residual row.
The matrix. A column per slot, each reading beside its difference from A, and a Residual row.
Slot A drawn as a ghost behind the slot in front, with Ghost on in the Compare task bar.
Ghost. Slot A behind the slot in front, so level and timing differences read at a glance.

The residual is the answer

What changed becomes its own waveform. The residual is A minus the slot in front, sample by sample. Once level and timing are matched, everything the two versions share cancels, and what is left is the change itself: on the waveform, where it shows where in the file the difference lives and how loud it is, and in the monitors.

slot A the reference slots B to F up to five takes auto-align to the sample level-match LUFS · polarity A − X the residual slot A the reference slots B to F up to five takes auto-align to the sample level-match integrated LUFS · polarity A − X the residual
How the residual is made. Takes align to the sample and match by integrated LUFS before the subtraction, so only what changed survives it.

Polarity matters for the subtraction. Invert polarity in the slot card flips a slot. It catches a wiring error, and in residual work it is the switch that lets two versions cancel.

Or read the numbers. The matrix puts every slot's readings side by side: integrated loudness, true peak, sample peak, LRA, DR, max momentary and short-term and the length, plus the dialogue loudness when a slot holds speech or the primary delivery target judges it, each beside its difference from A. Ghost (Compare › Show A Behind Other Slots) draws slot A behind the slot in front instead of subtracting it.

Against the encode, and the plugins

The streamed version. Under Music streaming, Analyze › Compare with the Apple Music Encode puts Apple Music's AAC encode of the master in a slot beside it: the matrix gives the encode's true peak beside the master's, and its residual plays what the encoder took away. Mastering

Before and after the processing. In the Edit task, the plugin chain's Render to Compare Slot renders the processed file into a new slot instead of the file, for an A/B of the chain. Editing

From the terminal. specula compare takes two files, level-matches them on request and writes the residual, resampling the second to the first file's rate. Batch and automation

Settle the comparison

"Better or just louder?" has an answer that plays. With the versions level-matched and aligned, the residual is the change itself, on screen as a waveform and in the monitors as audio, and the matrix gives the numbers from the same comparison.