Term	Meaning
ADSR	Attack, decay, sustain, release: the four stages of a note's volume envelope.
Agent	A program, often driven by an AI model, that calls ferraus commands or MCP tools on its own. See MCP.
Attack / release	How fast a compressor, gate, or limiter reacts when the level rises (attack) and recovers when it falls (release).
Bit depth	Bits per sample in integer audio (16, 24); ferraus processes in 32-bit float internally.
Bleed	Sound from one instrument that leaks into another's stem after separation (cymbals in the vocal stem). See Stems.
Chroma	Energy per pitch class (C, C#, ... B), used for key and chord detection.
CLAP, VST3, AU, LV2	Audio plugin formats; CLAP and VST3 are hosted in v1.
Crest factor	Peak level minus RMS level in dB; high means punchy, low means dense or heavily limited.
dBFS	Decibels relative to digital full scale; 0 dBFS is the loudest sample value.
dBTP	Decibels true peak, measured after oversampling so peaks between samples are caught.
De-esser	A compressor that reacts only to sibilance ("s", "sh") around 5–9 kHz.
Dither	Low-level noise added when reducing bit depth so quantization error becomes benign hiss.
Envelope	A value (gain, pan, pitch) that changes over time, set by keyframes.
Execution provider	An ONNX Runtime backend (CPU, CUDA, CoreML, DirectML).
Formant	A resonance of the vocal tract; keeping formants fixed keeps a voice sounding like itself when its pitch changes.
fx stage	One effect inside an fx chain or pipeline, written name:key=value.
Gate	Silences a signal while it is below a threshold.
Groove	A template of timing and velocity offsets per grid position, extracted from a performance.
Hum	A low buzz at the mains frequency (50 or 60 Hz) and its harmonics, picked up from power lines. Removed by dehum.
Integrated loudness	The loudness of a whole program, gated to ignore silence, in LUFS.
Key	The tonal center and mode of a piece (G major, E minor).
Lead sheet	Melody on a treble staff with chord symbols above and lyrics below.
Leveler	A slow automatic gain that raises quiet passages and lowers loud ones, like riding a fader. See Clean a podcast.
Limiter	A compressor with a very high ratio that keeps peaks under a ceiling.
LUFS, LU	Loudness units per EBU R128; LUFS is absolute, LU is a difference.
Mastering	Final processing of a mix for delivery: tonal balance, dynamics, stereo image, loudness, and format.
MCP	Model Context Protocol: the interface through which agents call ferraus as tools.
Mid/side	Stereo expressed as center (mid) and difference (side); the basis of widening and center extraction.
MIDI	A file of note and control events rather than audio.
Musical noise	Chirping, watery artefacts left by aggressive noise reduction. denoise --artifact-control reduces it.
MusicXML	The open interchange format for notation; ferraus's hub format between audio, MIDI, and engraved scores.
Noise profile	Per-frequency statistics of a noise-only region, used by spectral gating.
Null test	Subtracting two signals to check they are identical; a processed file at 0% mix must null against its input.
OMR	Optical music recognition: reading printed or scanned sheet music into a machine-readable score.
Onset	The moment a note starts. Transcription splits notes at onsets; --onset-sensitivity sets how readily.
Pan law	How gains are split between speakers as a source moves, so loudness stays even. See pan laws.
Pipeline	A YAML or JSON file describing a graph of stages, run with ferraus pipeline run.
PPQ	Pulses (ticks) per quarter note: the time resolution of a MIDI file.
Quantize	Moving MIDI note timings onto a rhythmic grid.
Relative minor	The minor key that shares a major key's notes (A minor for C major); key detection often confuses the two.
Retune speed	How fast pitch correction pulls toward the target note; 0 ms is the robotic effect.
RMS	Root mean square: an average level that tracks perceived loudness better than peaks.
SDR	Signal-to-distortion ratio in dB; the standard separation quality metric (higher is better).
SF2, SFZ	Sample-based instrument formats (SoundFont 2 and the SFZ text format).
SI-SDR	Scale-invariant SDR, used to evaluate denoising and in ferraus compare.
Sidecar	The <output>.ferraus.json provenance file written beside each output.
Solfa	Tonic sol-fa notation (do re mi, written d r m f s l t) relative to the key; readable without a staff.
Sound folder	A folder of SoundFonts, SFZ libraries, impulse responses, or plugins passed with --sound-path. See Sound folders.
Spectral gating	Attenuating time-frequency bins that fall below a noise threshold.
Stem	One instrument group separated from a mix (vocals, drums, bass, other, ...).
STFT	Short-time Fourier transform: the spectrum of short overlapping windows of audio.
TD-PSOLA	Time-domain pitch-synchronous overlap-add: pitch shifting that preserves formants.
Tempo map	The tempo (and changes) of a piece, used to convert beats to seconds.
Tier	A named set of separation models trading speed for quality (fast, balanced, quality, max). See Stems.
Transient	The sharp start of a sound (a drum hit, a pluck).
True peak	See dBTP.
TTA	Test-time augmentation: running a model on shifted copies and averaging, for better separation.
Varispeed	Changing speed and pitch together, like a tape played faster (speed); stretch and pitch change one without the other.