Analysis, References, and Song Intelligence

What can be measured

• The whole mix — integrated and short-term loudness (LUFS), true peak (dBTP), stereo correlation, spectral tilt, crest factor, eight-band tonal balance; the perceptual composites harshness, mud, and clarity; mastering QC — PLR/PSR (low PSR flags over-limiting), per-band correlation with a mono-compatibility flag, narrow resonances worth notching, and over-compression/over-limiting telemetry.
• Any stem, bus, or return — loudness, peak, band energy, plus its applied chain and gain-reduction telemetry.
• Masking — where one stem crowds another, per band, including whether panning already separates them.
• Sections — detected song sections with per-section loudness/tone/width ("does the chorus lift?").
• Phrases — phrase-level outliers on a stem: level dips, sibilant bursts, intonation (feeds ranged vocal rides and de-essing).
• Notes — per-note intonation on a monophonic stem, worst first (feeds Tune Notes).
• Chords & timing — chords detected from audio (with honest confidence), and rushing/dragging vs the grid (bias, jitter, swing).
• Separation artifacts — shimmer, transient smear, spectral holes, leakage on AI-split stems (feeds AI cleanup).

The song chips

The transport's LCD cells are the always-on readouts: BPM (detected tempo seeds the project grid once, when confident and only while the song still sits at the fresh-project 120 — the feed says "grid aligned to the detected tempo …" when it happens, and any tempo you set yourself, by cell, wheel or the agent, switches that seeding off for good, and the project remembers that. "Confident" includes the octave: a beat and its half or double are both periods of the same music, so when the detector cannot tell 170 from 85 it still tells the engineer what it heard but leaves your grid alone — typical above ~160 BPM, where you set the tempo yourself. Material with no rhythm — a pad, a podcast, a mix whose chorus is simply louder — is never given a tempo), KEY (detected key and mode), and SIG.

Tuning detection rides the same analysis — a detuned stem shows up in the engineer's observations, and whole-song or per-note correction is a sentence away.

Reference tracks

REFERENCE (the composer's status line, or + attach ▸ Reference) opens the Reference card:

• Load… a track you admire (any common format). The card profiles it and shows its loudness, the gap to yours, the biggest difference called out in words, and an eight-band tonal-balance chart of reference-minus-mix (a bar up = your mix needs more there).
• Match turns that comparison into a concrete numeric brief and hands it to the engineer to run. The reference's audio never enters your mix — only its measurements.
• Want more than one reference? Ask in chat: the engineer can profile multiple weighted references (weighting each for tonality, width, or loudness) and compare section-aware ("match the choruses to reference A").

Key tools in one place

• Match Key (Audio Clip View, Loop Deck pads) — pitch a clip/loop to the song's key once; it then stays where it was put.
• Follow Song Key (clip.auto_pitch) — the standing version of the same idea: the clip re-pitches itself every time the key track changes, by the nearest interval. Tell it what key the clip itself is in the first time (its root note) — without that there is nothing to move it from, and Zammoth says so rather than switching on and doing nothing. It adds to any fixed transpose you have dialled in rather than replacing it.
• KEY cell → transpose — move the whole song to a new key (mode-preserving).
• Tune Notes (clip menu) — per-note pitch correction on a monophonic take.
• Split into stems (AI) (track menu) — local 2- or 4-stem separation (vocals/drums/bass/other), staged progress, minutes of background work; the engineer can then measure and repair the artifacts.

Deep knowledge

The engineer carries a working engineer's knowledge — band maps, instrument landmarks, symptom → cause → tool tables, delivery-loudness norms, music theory — and can pull deeper reference material on demand (one topic at a time) when a question needs it: ask why, not just what, and you'll get the reasoning.