Glossary

Aliasing is what happens when a process generates frequencies above the Nyquist limit, half the sample rate, and they fold back into the audible range as inharmonic tones. At 44.1 kHz, a 30 kHz harmonic from a saturator reflects off the 22.05 kHz ceiling and lands at 14.1 kHz, unrelated to the note you played. Distortion and clipping plugins are the usual offenders, which is why many offer oversampling.

Attack is the time a compressor takes to ramp its gain reduction to full depth once the signal crosses the threshold. Reduction starts the instant the threshold is crossed; the attack setting only controls how fast it deepens. Drums are where it shows: the snap of a snare lives in roughly its first 10 to 30 ms, so a 1 ms attack flattens it while a 30 ms attack lets it pass before the clamp arrives.

Automation is a recorded curve that moves a parameter over time, so a fader ride or filter sweep plays back identically on every pass. Order of operations is the part to watch: volume automation on a channel is post-insert in most DAWs, so it will not change how hard you drive a compressor sitting on that channel. To automate the level feeding your inserts, automate a gain plugin placed before them.

Bit depth is the number of bits stored per sample, which sets the gap between the loudest possible level and the noise floor at roughly 6 dB per bit. 16-bit gives about 96 dB of range, 24-bit about 144 dB on paper. The extra bits buy headroom rather than finer detail; you can track around -18 dBFS and the noise floor stays irrelevant. When exporting to 16-bit, dither so quantization error becomes benign noise instead of distortion.

Buffer size is how many samples your interface processes per block, and it sets the latency you feel: 256 samples at 44.1 kHz is about 5.8 ms each way, before converter and plugin delay add more. Round-trip figures are what matter for monitoring, usually double the buffer plus a couple of ms. Track at 64 or 128, then raise it to 512 or 1024 for mixing, when latency stops mattering and CPU headroom does.

A bus is a channel that sums the output of several tracks so you can process or level them together, a drum bus and a vocal bus being the usual cases. Buses and sends get conflated: routing a track's output to a bus moves the whole signal, while a send takes an adjustable copy and leaves the original in place. Whether that copy follows the fader is the pre and post fader question.

A chorus is a delay line whose delay time is being moved by an LFO, mixed back against the dry signal. Because the delay is changing rather than fixed, the moving copy is also pitch shifted: a shrinking delay plays sharp, a growing one flat, and the amount depends on how fast the delay is moving rather than how far it has moved. At 2 ms of depth and 1 Hz that works out to about 21.6 cents sharp at the steepest point.

Clipping reshapes a waveform wherever it crosses a boundary, flattening everything above it. It is a transfer function with no memory, which is what separates it from a limiter: a limiter changes gain over time and can begin reducing before the peak arrives. Both can leave a peak meter on the same number. On one test transient the clipped output had a flat top twenty samples long and the limited one a single sample.

Comb filtering is the row of evenly spaced notches you get when a signal sums with a delayed copy of itself. The spacing follows the delay: a 1 ms offset puts the first null at 500 Hz and repeats every 1000 Hz above it. It is why two mics on one source, or a duplicated track nudged a few samples, can hollow out a tone. Doubling the delay halves the spacing, chewing lower into the spectrum.

Components are sold to a tolerance, and the tolerance is the whole specification: a resistor marked 1% is guaranteed only to be within 1%, and film capacitors are commonly 5 or 10%. A filter corner depends on those values and inherits their error, so the two channels of a piece of hardware are never quite the same as each other. On one modelled pair the channels sat 6.6 microseconds apart, correlated at 0.99913, with a difference signal 33.6 dB below the centre.

A correlation meter reports how similar the left and right channels are in shape, which is not the same question as how wide the image is. It runs from +1, where the channels are identical, through 0, down to -1, where they are identical with opposite polarity. It is level independent, so a signal panned off centre still reads +1: panning changes the balance between the channels, not the shape of either one.

Crest factor is the gap between a signal's peak level and its RMS level, expressed in dB. A sine wave measures 3 dB; raw multitracked drums can run 18 dB or more, while a heavily limited master might sit near 6. It is the number compression and limiting actually change, which makes it useful for judging how squashed a mix already is before you reach for another limiter. The full arithmetic, and why matching peaks does not make a fair test, is in the crest factor guide.

Cutoff frequency is the point where a filter's attenuation reaches 3 dB, the conventional marker for the edge of the passband. A low-pass set to 8 kHz is already 3 dB down at 8 kHz itself, then keeps falling at its slope, typically 12 or 24 dB per octave. That means the filter is audibly working below the number on the dial, and with resonance up it boosts right at the cutoff before the drop.

dBFS is decibels relative to full scale, the highest value a digital system can represent, so 0 dBFS is the ceiling and every real level reads negative. Producers trip on the analog crossover: many hardware emulation plugins calibrate their internal 0 VU to -18 dBFS (some use -20), so a signal peaking near 0 dBFS drives the model far harder than the original unit was ever run. Keep the average around -18 and the emulation behaves.

DC offset is a constant shift that moves the whole waveform off the zero line, effectively a 0 Hz component riding under the audio. It eats headroom silently, since peaks on one side hit the ceiling early, and it puts a click at every edit point because cuts no longer land on zero. Some synth waveforms and cheap interfaces introduce it; a high-pass around 10 to 20 Hz removes it without touching anything audible. How to measure it, and why it is not the same as waveform asymmetry, is in the DC offset guide.

A de-esser is a compressor that reacts only to sibilance, the s and t energy sitting between roughly 5 and 8 kHz on most voices. Cut too deep and the esses become a lisp; solo the detection band to find where the ess actually lives, then pull only a few dB. Split-band designs duck just that band, while wideband designs drop the whole vocal level on every ess, dulling whatever else is happening at that moment.

Decay is the time an envelope takes to fall from its peak to the sustain level once the attack has finished. It is a duration, and the level it falls to belongs to sustain rather than to decay. That coupling has a consequence worth knowing: with sustain at maximum there is nothing for the decay stage to travel, so changing it produces an identical envelope and the control appears to do nothing.

Dither is low-level noise added on purpose before a reduction in bit depth. Without it, truncating to 16 bits rounds each sample to the nearest step, and the rounding error correlates with the signal, which you hear as distortion on fades and reverb tails. Dither trades that for a steady hiss near -96 dBFS, below anything a playback system will expose. Apply it once, at the final bounce, and never before further processing.

Dry/wet is the balance between an effect's unprocessed input (dry) and its processed output (wet). On an insert, the knob does what it says. On a send, set it to 100 percent wet, because the dry signal already exists on the source channel; anything less doubles the dry copy through the return bus and shifts the level every time you touch the send. Parallel compression is the same idea driven from the wet side.

A dynamic EQ is a filter whose gain is driven by a detector instead of being held where you set it. It is not a multiband compressor: there are no crossovers and no need to divide the whole spectrum into contiguous bands, so one narrow bell can move while everything around it stays untouched. FabFilter band limits its detector to the band own frequency range by default, and allows a dynamic range of -30 to +30 dB.

An envelope is a control signal that reshapes a parameter over time, retriggered by each note; ADSR names its four stages: attack, decay, sustain, release. Sustain is the odd one out: a level, not a time. The other three are durations, but sustain is the height the envelope holds while the key stays down. That is why lengthening decay does nothing audible when sustain sits at full, and why a pluck wants sustain at zero with decay doing the work.

An envelope follower tracks the loudness contour of an audio signal and turns it into a control signal, which is how auto-wahs and a lot of ducking effects work. Release is the control to watch. One cycle of an 80 Hz bass lasts 12.5 ms, so a release faster than that makes the follower ride individual waveform cycles instead of the note, and whatever it modulates buzzes at the pitch of the bass. Slow the release until the ripple stops.

ERB, equivalent rectangular bandwidth, is a measure of how wide the ear's own frequency bands are at a given frequency. They are narrow in the bass and widen as frequency rises: on the standard Glasberg and Moore formula, about 35 Hz wide at 100 Hz and about 133 Hz wide at 1 kHz. Processing in ERB bands rather than equal-width FFT bins puts the resolution where hearing has it, which is why spectral reverbs and key-aware EQs mention it.

A Euclidean rhythm distributes a number of hits as evenly as possible across a number of steps, which produces patterns without drawing them. Five hits across sixteen steps cannot be spaced evenly, so the gaps alternate between the two nearest whole numbers: 3, 3, 4, 3, 3. Rotation moves where the pattern begins without changing either count, and is a separate control from both.

The proportion of a delay's output routed back into its own input, which turns a single echo into a tail. It multiplies rather than adds, so the knob is far from linear: 50 percent loses 6.02 dB on every repeat and reaches 60 dB down in ten of them, while 90 percent loses only 0.92 dB and needs about 66. See the delay feedback guide.

The fast Fourier transform is the algorithm that turns a slice of audio into a spectrum, and it sits under every analyzer and most spectral tools. Its resolution is a trade. A 1024-sample window at 44.1 kHz gives frequency bins about 43 Hz wide, which cannot tell two low bass notes apart, since whole octaves down there span less than that. A longer window sharpens frequency and blurs timing, which is why spectrograms of drums look smeared at big FFT sizes. Details live in the DSP section.

Filter slope is how quickly a filter attenuates past its cutoff, measured in dB per octave; each pole contributes 6 dB per octave, so a 4-pole filter falls at 24. Slope describes what happens well beyond cutoff, not at it. At the cutoff frequency itself a standard design is only 3 dB down, whatever the slope, so a 24 dB per octave high-pass at 100 Hz still lets plenty of 90 Hz through. The full guide covers where each slope earns its keep.

A formant is a resonant peak in a sound's spectrum whose position is set by the body that made the sound, not by the note being played. Vowels are formant patterns; an adult "ah" puts its first two peaks near 700 and 1100 Hz whatever the pitch. This is why plain pitch shifting sounds wrong on vocals: it drags the formants along with the pitch, shrinking or enlarging the apparent singer. Formant-preserving shifters move the note and leave the peaks where they were.

FM synthesis modulates one oscillator frequency with another, putting sidebands at the carrier plus and minus whole multiples of the modulator. The timbre is decided by the ratio between the two rather than by either frequency on its own, which is why the control is usually labelled Ratio. A modulator tuned to a ratio follows the keyboard; one fixed in hertz does not, so a modulator at 400 Hz gives a ratio of 2.00 at a 200 Hz note and 1.33 at 300 Hz.

A frequency shifter moves every component of a signal by the same number of hertz, which is an absolute distance rather than a musical interval. One component goes in and one comes out: 440 Hz shifted up by 100 Hz gives 540 Hz and nothing else. On a harmonic signal the gaps between partials survive, because adding a constant does not change the differences, but the ratios do not, so the result stops reading as a note.

The Gabor limit is the trade-off between knowing exactly when a sound happens and knowing exactly what pitch it is. The shorter a sound, the wider the spread of frequencies it has to contain, so a 1 ms click cannot have a precise pitch and a precise pitch needs time to establish itself. It is why very short grains in a granular synth sound like clicks rather than notes.

Gain staging is setting the level at each point in the chain so every stage receives what it was designed for. Inside a modern DAW the mixer is 32-bit float and will not clip between plugins, so the practice is really about the plugins themselves. Analog-modeled processors are usually calibrated so that around -18 dBFS behaves like the hardware's nominal level; feed them 10 dB hotter and you get more drive than the preset designer intended. Level matters when comparing too, so match outputs before judging.

The Haas effect is the ear fusing two arrivals of the same sound into one when the gap is under roughly 35 ms, placing the source at whichever copy arrived first. Pan a dry copy left and a 15 ms delayed copy right and you get width with no level change. The bill arrives in mono: summed, that pair comb filters, with notches every 67 Hz or so for a 15 ms delay. Check mono before committing, or read what nudging does to phase.

Hard sync resets one oscillator phase every time another completes a cycle. The reset happens on the master schedule, so the output repeats at the master rate and the pitch stays where the keyboard put it, while the interrupted waveform changes the harmonic content. The bright spot in the spectrum follows the slave: at a 100 Hz note with the slave at 430 Hz, the loudest partial above the fundamental sits at 400 Hz. Whole-number ratios do nothing, because the reset then arrives exactly when the slave was going to restart anyway.

Harmonics are frequency components at whole-number multiples of a fundamental; a 100 Hz note carries them at 200 and 300 Hz and at every multiple beyond. Their musical positions explain saturation character: the 2nd harmonic lands an octave above the note, the 3rd an octave plus a fifth, so a device favoring even multiples reinforces the note itself while odd multiples add intervals the chord may not want. Which devices favor which is the subject of even and odd harmonics.

Headroom is the gap between a signal's peaks and the point where the system clips. Digital's trap is that the meter is not the whole story: sample peaks at 0 dBFS can reconstruct above full scale between samples, and lossy encoders push overs further, which is why mastering for streaming usually leaves about 1 dB of true-peak headroom (a ceiling of -1 dBTP). Headroom while mixing is cheaper still, since a 32-bit float bus can take absurd levels; the converter and the codec cannot.

A high-pass filter passes frequencies above its cutoff and attenuates those below it. The number on the dial misleads: cutoff marks the -3 dB point, not where the low end disappears. A 12 dB per octave high-pass set at 100 Hz leaves 50 Hz only about 12 dB down, still very present on a club system. Steepness is a choice with side effects of its own, covered in filter slopes explained, so check an octave below cutoff before deciding the rumble is gone.

Hysteresis gives a gate two thresholds instead of one: a higher level to open and a lower level to close. Between them the gate keeps whatever state it already had, so a detector hovering near the opening level cannot flip it repeatedly. Ableton calls the control Return. Set the opening threshold at -24 dBFS with a 6 dB return and the gate will not begin closing until the detector falls below -30 dBFS. See the gate hysteresis guide.

Intermodulation is what a nonlinear process produces when more than one frequency reaches it at once: sums and differences of the inputs and their multiples, rather than harmonics of any single one. The direction is what matters. A harmonic can only appear above the frequency that made it, but a difference tone can land below everything you put in. A 50 Hz sub and a 130 Hz bass through one saturator produce a component at 30 Hz, from twice the sub minus the bass.

Knee is how gradually a compressor moves from no gain reduction to its full ratio around the threshold. A hard knee switches at the threshold exactly; a soft knee spreads the transition across a stated range, so a 6 dB knee starts compressing gently 3 dB below the threshold and only reaches the full ratio 3 dB above it. That is why soft-knee settings show gain reduction earlier than the threshold reading suggests.

Latency is the delay between audio entering a system and leaving it. Buffer size sets most of it: 256 samples at 44.1 kHz is 5.8 ms each way, before the converters add their bit. Plugin delay compensation only fixes playback. A lookahead limiter or linear-phase EQ on the master adds its delay to everything you monitor through the DAW while tracking, so bypass those, or record at a small buffer and mix at a big one.

Low-frequency oscillator: an oscillator cycling below the audible range, usually under 20 Hz, used to modulate other parameters rather than be heard directly. Retrigger is the control worth finding. A free-running LFO sits at a random point in its cycle when each note starts, so a filter sweep lands differently on every hit. Key-synced mode restarts the cycle per note, which is what you want for repeatable wobbles and bass movement.

A compressor with a ratio of roughly 10:1 or higher and an attack fast enough that nothing passes the ceiling; modern brickwall designs are effectively infinite ratio. True peak and sample peak are not the same number. Lossy encoding can reconstruct inter-sample peaks above your sample-peak ceiling, which is why streaming services recommend a true-peak ceiling of -1 dBTP rather than pushing right up to 0 dBFS.

A linear-phase filter delays every frequency by the same amount, so it changes the level of each frequency without moving any of them in time relative to the others. The costs are latency, often tens to hundreds of milliseconds for low-frequency work, and pre-ring. It is the right tool when an EQ'd signal will be summed with another version of itself, as in parallel chains, multi-mic sources and mid/side EQ.

A delay on a dynamics processor's audio path so the detector reads the signal early and starts reducing gain before a peak arrives, usually 1 to 5 ms ahead. Lookahead is real latency: a limiter with 2 ms of lookahead reports at least 2 ms to the DAW. That is harmless on a master bus and a problem while tracking, which is why a limiter's zero-latency mode usually just switches lookahead off.

Loudness Units relative to Full Scale, the K-weighted loudness measure from ITU-R BS.1770 that streaming services use for normalization. Integrated LUFS averages the whole track, while short-term reads a 3 second window. Mastering hot anyway buys nothing: Spotify turns a -8 LUFS master down to -14, so the crushed transients stay and the level advantage disappears. Broadcast is a different world, targeting -23 LUFS under EBU R128.

The output gain on a compressor that restores the level lost to gain reduction. It is also the main way compressors flatter themselves: louder reads as better even at differences under 1 dB, so a plugin adding 4 dB of makeup wins any bypass comparison regardless of what the compression did. Match bypassed and processed levels before judging, and treat auto-makeup as a rough guess, because most implementations overshoot.

A stereo encoding that stores sum and difference instead of left and right: mid is (L+R)/2, side is (L-R)/2. The trap is treating the side channel as the wide stuff and boosting it freely, because everything in side cancels when the mix folds to mono, which keeps only mid. Before reaching for a widener, check what each channel actually contains, since side often holds reverb tails and bleed rather than the parts you meant to widen.

Minimum phase is how ordinary analogue and digital EQs behave: no latency, and all of the filter's ringing happens after the input that caused it, with frequencies around the band shifted slightly later than the rest. On a single track this is rarely audible. It becomes audible when the filtered signal is blended with an unfiltered copy, where the timing shift causes partial cancellation.

How much of a mix survives being summed to one channel, which still happens on phone speakers and plenty of club systems. Delay-based widening is the usual casualty: a 10 ms Haas offset spreads a mono source across the stereo field, then comb-filters into a thin version of itself when summed. Anything that exists only in the side channel vanishes outright in mono, so check by listening, not by trusting a correlation meter alone.

The rule a monophonic instrument uses to pick which held key controls its single voice: low, high or last note. Hold C3 and add G3 then G2, and low priority sounds C3 then G2, high priority sounds G3 throughout, and last priority follows each new key. Whether the envelopes restart is a separate setting. See the note priority guide.

Half the sample rate, and the highest frequency a digital system can represent: 22.05 kHz in a 44.1 kHz session. Anything a process generates above it folds back below it as aliasing, landing at frequencies with no musical relationship to the source. Saturation is the usual culprit, since it creates harmonics far above the fundamental: distort a 5 kHz tone and its seventh harmonic at 35 kHz reflects back down to 9.1 kHz.

Running a process at a multiple of the session sample rate, typically 2x to 16x, so the harmonics distortion creates land below the raised Nyquist point instead of folding back as aliasing. Two costs come with it: CPU scales roughly with the multiple, and the downsampling filter adds latency, especially in linear-phase mode. It pays off on clippers and saturators; a delay or a gain utility gets nothing from it.

The attenuation a mixer applies at center pan so a signal keeps the same perceived loudness anywhere in the stereo field. The common center cut is 3 dB, with DAW options running from 2.5 to 6 dB. The mistake is changing it mid-project: every centered track shifts level relative to panned ones, which silently rebalances a mix you thought was done. Pick the law before mixing starts and then leave it alone. Where the 3 dB figure comes from, and how to measure which law a DAW is using, is in the pan law guide.

Parallel compression sums a compressed copy of a signal with the original, so the blend control sets the balance between two pieces of audio rather than between two gain curves. The trap arrives before the compressor has done anything: two identical time-aligned copies sum 6.02 dB louder than one, so a return passing dry signal alongside compressed signal is already wrong. Use a fully wet aux or the plugin own mix control, not both at once.

A partial is one sine component of a sound. In a harmonic sound such as a sawtooth or a sung note, the partials sit at whole-number multiples of the fundamental and are called harmonics. In a bell, a drum or a struck plate they do not, which is why those sounds have a less certain pitch. Additive synthesizers build a sound by setting the frequency, level and envelope of each partial.

The phantom centre is the image that appears between two speakers when the same sound comes from both at equal level. It is not the same as the mid channel: mid is left plus right, so it also carries part of every source panned partly to one side, while the phantom centre is only what is identical in both. Tools that claim to isolate the centre are trying to separate the two.

The position of a waveform within its cycle, measured in degrees, where 360 degrees is one full cycle. The standard confusion: the polarity button flips the waveform upside down, which is not a phase shift, although it matches a 180 degree shift on a lone sine wave. Real-world phase problems are timing offsets. Two mics on one snare a few centimetres apart produce comb filtering, and no polarity flip fully repairs that; alignment does.

Noise with equal energy per octave, which means its spectrum falls 3 dB per octave as frequency rises. White noise carries equal energy per hertz and sounds far brighter by comparison. Full mixes distribute energy in a roughly pink way, which is why pink noise works as a balancing reference: bring each track up against a fixed pink noise bed until it just pokes through, and you get a usable rough balance in minutes.

The number of voices an instrument can sound at once. Voices get used up faster than notes: a four-note chord on a patch with 8-voice unison eats 32 voices, and every release tail holds its voice until it fully fades. When the count runs out the synth steals the oldest voice, which is what that clipped-off pad tail actually is. Raise the voice limit or shorten the release before blaming the preset.

The gap a reverb leaves between the dry signal and the onset of its reflections. Set 20 to 40 ms on a vocal and the consonants land before the tail arrives, which keeps the voice in front of the reverb instead of buried inside it. Left at zero on small rooms, the reverb piles onto the transient itself and reads as smear rather than space.

Pre-ring is filter ringing that starts before a transient instead of after it. Linear-phase filters produce it because their response is symmetric in time. It is worst for narrow, low-frequency moves: a +10 dB boost at 55 Hz with a Q of 8 on a kick is already above -30 dB about 120 ms before the hit. After silence it is exposed; inside a busy mix it is mostly masked.

Q is the ratio of a filter band's center frequency to its bandwidth, so a higher Q means a narrower band. A Q of 1.41 spans about one octave; a Q of 10 centered at 200 Hz grabs a band only 20 Hz wide. Many EQs use proportional Q, which widens the band at low gain settings, so the same number behaves differently across plugins. On a resonant filter, higher Q boosts the region around the cutoff instead.

Ratio is how strongly a compressor reduces signal above the threshold, written as input change against output change. At 4:1, a peak that goes 8 dB over the threshold comes out 2 dB over. From roughly 10:1 upward you are limiting. The panel number is a maximum, though: with a soft knee the effective ratio ramps up gradually around the threshold, so a 4:1 driven only lightly into the signal often behaves closer to 2:1. The transfer equation and worked examples are in the compressor ratio guide.

Release is how long a processor takes to let go: on a synth envelope, the fade after note-off; on a compressor, the time gain takes to recover once the signal drops back below the threshold. The compressor version is the one that bites. Set it shorter than one cycle of your lowest frequency and the gain reduction starts tracking the waveform itself, which is distortion. A 40 Hz bass repeats every 25 ms, so releases under that will chew on it.

Ring modulation multiplies two signals together and outputs the sum and difference of the frequencies in each. One input component becomes two: a 440 Hz sine against a 100 Hz carrier gives 340 Hz and 540 Hz, and the original 440 Hz is not part of the result. On a signal with several partials the component count doubles. Below roughly 20 Hz the same multiplication is heard as tremolo rather than as separate tones, which is the only thing separating the two effects.

RMS, root mean square, squares every sample in a measurement window, takes the mean of those squares and then the square root. It describes average level rather than the highest single sample, which is why a brief spike moves a peak meter far more than an RMS one. Window length is part of the measurement, so two meters can disagree on the same audio. Peak minus RMS gives crest factor.

Round-robin is a sampler mode that cycles through several alternate recordings of the same note, so repeated hits do not replay one identical file. A single snare sample fired on every eighth note is the machine-gun effect; even two round robins break it, and drum libraries commonly record 4 to 10 per velocity layer. Watch for samplers that reset or randomize the cycle position, because a bounce can then start on a different variation than the playback you approved.

Sample rate is how many amplitude snapshots per second a digital system stores, and it caps the highest frequency you can represent at half that figure, the Nyquist limit. 44.1 kHz captures up to 22.05 kHz. Raising the rate does not add detail you can hear; the practical benefit is headroom for saturation and other nonlinear plugins, whose new harmonics would otherwise land past Nyquist and fold back down as aliasing. Oversampling inside the plugin gets you that without a 96 kHz project.

Saturation is mild nonlinear distortion: the waveform gets squashed and new harmonics appear at whole-number multiples of the input frequencies. Symmetrical clipping adds odd harmonics, asymmetrical circuits add even ones as well, and the second harmonic sits exactly one octave above the note. The reliable trap is level. Saturation adds energy and top end, so the processed signal is louder, and louder reads as better. Match output level to input before you judge it.

A sidechain is the detector input of a dynamics processor, split from the audio the processor actually touches. Route the kick into a compressor on the bass and the bass ducks on every kick hit. One step gets skipped: high-pass the sidechain around 60 to 100 Hz on bus compression, so low end stops driving the gain reduction and pumping the whole mix. For rhythmic ducking, a volume shaper is often the better tool than a compressor.

Stereo linking decides how much a stereo dynamics processor lets one channel's level control the gain applied to the other. FabFilter puts the ends plainly: at 0% the channels operate independently, at 100% they receive the same gain reduction. Steinberg describes the same thing from the detector side, using the highest-level channel for analysis when linking is on. It does not change how hard the processor works, only who else is turned down: on one measured example a burst on the left alone moved the stereo balance by 10.5 dB unlinked and not at all fully linked.

Sustain is the level an envelope holds while the key stays down, and it is the only one of the four ADSR controls that is not a time. Turning it down does not make a note shorter, it makes the held part quieter. With attack, decay and release left alone, an envelope at sustain 0.85 and the same envelope at 0.30 rise, fall and fall silent at exactly the same moments.

Threshold is the level at which a dynamics processor starts working: above it a compressor turns gain down, below it a gate closes. There are two catches. A soft knee begins compressing below the stated threshold, so a 6 dB knee starts 3 dB early. And lowering the threshold plus adding makeup gain makes everything louder, which skews any bypass comparison toward the processed version. Match levels before deciding the compressor sounds better.

A transient is the short burst of energy at the onset of a sound, the first few milliseconds of a snare crack or a plucked string. Transients carry the peaks: a raw drum bus can peak 15 dB or more above its average level, and that gap is what compressors and clippers spend. Attack times under about 5 ms clamp the transient itself, while 10 to 30 ms lets it pass and squeezes what follows. How a shaper detects one, using two envelope followers rather than a threshold, is in the transient shaper guide.

True peak is the estimated level of the reconstructed analog waveform, measured in dBTP rather than dBFS. Sample values can all sit at or below 0 dBFS while the curve between them swings higher, so a master that never touches the clip light can still overload a converter or a lossy encoder. Meters estimate it by oversampling, 4x in the BS.1770 spec, and EBU R128 puts the ceiling at -1 dBTP.

Unison stacks a synth voice into several detuned copies per note, usually spread across the stereo field; the classic supersaw is 7 sawtooths. It costs you twice. Polyphony multiplies, so 7-voice unison under a 4-note chord is 28 voices. And the stereo spread is fragile in mono, where detuned copies can cancel instead of reinforcing. Solo the mono fold-down before committing a unison bass, and consider keeping the lowest octave a single voice.

A velocity layer is a set of samples mapped to one slice of the MIDI velocity range, so harder key strikes trigger recordings of harder playing rather than the same file turned up. MIDI offers 127 usable velocity values; a 4-layer piano switches samples roughly every 32 of them, and the seam is audible when adjacent layers were recorded at mismatched levels. Crossfading between layers hides the jump at the cost of a slightly softened attack. Why a one-step velocity edit can switch layers, and what that does to a random-velocity setting, is in the MIDI velocity guide.

A wavetable is a stored row of single-cycle waveforms that an oscillator reads through, with a position control for scanning between frames; Serum's tables hold up to 256 frames of 2048 samples each. That scanning is what separates it from a plain oscillator. Aliasing is the limit: bright frames played high on the keyboard push harmonics past Nyquist unless the synth band-limits playback per octave, which is why cheap wavetable oscillators fizz in the top range.

White noise is a random signal with equal power per hertz, a flat spectrum. Each octave spans twice the bandwidth of the one below it, so every octave carries 3 dB more energy than the last, which is why white noise reads as hiss piled at the top. Pink noise, falling 3 dB per octave, is the one that sounds even to human ears. Use white for hats and risers; reach for pink when checking mix balance.

A timing reference that makes digital audio devices agree on when each sample occurs. It is not MIDI clock, which carries musical tempo and transport and cannot synchronise sampling at all. A rig can need both at once: word clock between the interfaces, MIDI clock between the sequencers.