Murmur is a granular instrument you play from the keyboard. Drop any audio file on it and the sample is broken into grains you can scatter, stretch, tune, chop and lock to the hits it finds in the file. The cloud you see is the sound: every particle is a real grain leaving the engine.
It runs as a VST3 and an Audio Unit in your DAW, and as a standalone app. This manual documents version 2.6.0.
Quit your DAW, then run the .pkg and follow it through. You can pick which
pieces to install:
/Library/Audio/Plug-Ins/VST3/Murmur.vst3/Library/Audio/Plug-Ins/Components/Murmur.component (Audio Unit, for Logic
and GarageBand)/Applications/Murmur.app (the standalone)macOS 11 or later, Intel or Apple Silicon. Release builds are signed and notarized, so macOS opens them with no security prompts.
Quit your DAW first. Windows cannot replace a plug-in file while a DAW has
it open. Then either run Murmur-x.y.z-Windows-Setup.exe, which puts the VST3 in
C:\Program Files\Common Files\VST3\Murmur.vst3 and the standalone in
C:\Program Files\Sounds Like You\Murmur\, or take the plain zip and copy the whole
Murmur.vst3 folder (it is a bundle, not a single file) into
C:\Program Files\Common Files\VST3\ yourself.
VST3 and standalone only. There is no Audio Unit on Windows.
The ones you save:
macOS: ~/Library/Audio/Presets/Sounds Like You/Murmur/
Windows: %USERPROFILE%\Documents\Sounds Like You\Murmur\Presets\
The starter templates the installer puts down, which Murmur reads and never writes to:
macOS: /Library/Application Support/Sounds Like You/Murmur/Presets/
Windows: C:\ProgramData\Sounds Like You\Murmur\Presets\
macOS: delete whichever of the three paths above you installed. Windows: use Settings, Apps, Murmur, Uninstall, or delete the files you copied by hand. Your saved presets are left alone either way. To remove those too, delete the preset folder listed above.
Murmur checks for a newer version at most once a day, in the background, and shows a small amber UPDATE chip in the header when one is live. Click it to download. The new installer replaces the old version in place and your presets are untouched. The check is a single request that downloads one small file: the current version number, and the list of refunded licences described in section 2. The request carries no identifier, no licence data, no audio and nothing about your session, though like any web request it necessarily shows our server an IP address. If you are offline it fails quietly and you never see it.
Murmur is free to keep, and a licence key unlocks the rest of it. There is no trial, no countdown and no expiry date. The free version does not stop working, does not go quiet after a while, and does not degrade: it is a finished instrument that happens to be a subset of a larger one.
Read this table before you go hunting for a dimmed control. Nothing below is broken or faulty when it reads dim. It is telling you which half of the instrument you are in.
| Free, always | Full version adds |
|---|---|
| Loading any sample, by drag and drop or the file dialog | MORPH, and the whole second sample layer |
| POS, SCAN, DENSITY, SIZE, SHAPE, PITCH | COHERE |
| JITTER, SPREAD, MAGNET, SLICES | WALK play mode, and its step sequencer drawer |
| The ENV 1 envelope, including its slope curves | ENV 2 and ENV 3 |
| CLASSIC and SLICE play modes, editing slice boundaries on the waveform, the GRID divisions, and CLEAR / DETECT | REPEAT |
| HZ, SYNC and KEY grain clocks, and the sync division | G-FILTER and G-COMP, the two per-grain processors on the MOTION card's FILTER and COMP pages |
| TUNE, SCALE, VOICES, GLIDE, OUT and the meter | The FX rack, all seven modules, its bypass and its modulation destinations |
| DICE, the cloud colour, and loading any preset | Everything that modulates: LFO 1, LFO 2 (drawn shapes and grain sync included), RANDOM, GRAIN RND, the four macros, the source badges, the mod rings, the ROUTING matrix |
| The nine starter templates | Saving your own presets |

A paid control is never hidden and never removed. It sits exactly where it will be when you unlock it, dimmed, and it still answers a hover with a tooltip that says what it does. The three drawer tabs stay clickable in the free version for the same reason: the drawer opens and you can read it, with a dim veil over the top instead of a locked door.

This manual documents the whole instrument. Where a section covers something the full version adds, it says so at the top.
That is the whole instrument. Everything else changes how those grains are cut and where they land.
Two gestures worth learning straight away. Scroll over the display to zoom into the sample; a FIT pill appears in the top right corner showing the current magnification, and clicking it puts the whole file back in view. Scrolling the other way does the same thing. And double-click any knob to put it back to its default, anywhere in the plug-in, including the dials inside the FX rack.

Everything lives on one screen. The display is on the left, a column of cards on the right, a quiet strip of utilities under the display, nine dials across the middle, and a footer. Three drawers open over the lower half from the tabs at the bottom right.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21

A small gear sits at the right of the header, beside the cloud colour swatch. It is the one menu in Murmur that is about the plug-in rather than about the sound, which is why it is out of the way rather than on the panel:
Every dial here is a host parameter, so you can automate it or map it to a controller. Drag a dial up and down to move it, or use the scroll wheel. Double-click returns it to its default. Hover anything for a one line reminder.
| Control | Range | What it does |
|---|---|---|
| JITTER | 0 to 1 | Randomises each grain's start around the read point. Loosens a rigid loop into a scatter. The amber band on the waveform shows the range. |
| SPREAD | 0 to 1 | Stereo width. Pans grains left and right for a wide, diffuse cloud. |
| COHERE | 0 to 1 | Locks grain timing, position and length to the sample's own pitch, level-matched so turning it changes the character, not the volume. Turned up, a pitched sample plays like a solid synth tone instead of a texture. The lock scales with how confident the pitch analysis is — it reads the audio itself, never the filename — and with no usable pitch the dial dims and its tooltip says why. Full version. |
| SIZE | 10 to 500 ms | Length of each grain. Short is clicky and textural, long is smooth and pad-like. Default 50 ms. |
| DENSITY | 0.5 to 200 /s | Grains per second. Low is sparse and pointillist, high is a solid wash. Only in use when the clock above it reads HZ. |
| POS | 0 to 1 | Where in the sample grains are taken from. Dragging across the display moves this. Dimmed in SLICE mode, where the key you press is the position. |
| MORPH | 0 to 1 | How much of each grain comes from the second sound instead of the first. Inert until you load a second layer. Full version. |
| PITCH | ±24 st | Transposes every grain without changing grain length. |
| MAGNET | 0 to 1 | Pulls grain starts toward the slice boundaries on the waveform. At 0 the scatter is free, at full every grain lands exactly on a boundary. |
| Control | Range | What it does |
|---|---|---|
| SCAN | ±2 | Sets the read point moving on its own. 1.00 crawls forward at the sample's own speed, 2.00 is double, negative runs backwards, and it wraps at the ends. The centre is a detent, and parked there POS stays put. |
| SHAPE | 0 to 1 | The window each grain is heard through. Fully down is a smooth rounded fade, the centre is a flat top that only tapers the very edges, and fully up is a soft bell. See the chop recipe in section 11. |
| SLICES | 0 to 1 | Sensitivity of the hit detection that places slice boundaries. Low keeps only the strongest hits, high keeps everything. Goes inert while the boundaries are authored rather than detected, because then the count is no longer its decision, though it still guides RE-SLICE. See section 6. |
| GRID | TRANSIENT or a division | The combo under SLICES: where the boundaries come from. TRANSIENT is the detector; any note value lays them on the beat instead. See section 6. |
| TUNE | on / off | Remaps the keyboard to the sample's real key,
so middle C plays the note you expect. The readout beside it names the root Murmur found
and where it came from: file from the filename, a percentage from pitch
analysis, or manual if you set it. Click the readout to override it, pick an
octave, or put it back to AUTO. A dimmed readout means the sample is not strongly pitched
and TUNE has little to grab. |
| SCALE | 9 types | Snaps every grain's pitch to a scale: chromatic, major, minor, both pentatonics, dorian, mixolydian or lydian, from any root. Off leaves pitch untouched. Pitch bend still glides freely either way. |
| VOICES | MONO to 16 | Polyphony limit. A note past the cap steals the oldest voice. Fewer voices means less CPU. |
| GLIDE | 0 to 2000 ms | Appears only at MONO. How long the pitch takes to slide from the last note to the new one. All the way down is an instant jump. |
The strip above DENSITY decides what sets the grain rate.
| Mode | What it does |
|---|---|
| HZ | The DENSITY dial sets grains per second, free of the host. |
| SYNC | A musical division of the host tempo, from 1/2 down to 1/32 including 1/8 and 1/16 triplets. DENSITY goes inert. This is where REPEAT lives. |
| KEY | The pitch of the note you play sets the grain rate, which pushes it up into audio range and turns the stream itself into the tone. The grain train is now the oscillator, so you can play chromatically and stay in tune whatever key the sample itself is in. It wants short grains: at KEY rate the grain rate is the note frequency, so long grains overlap many to a cycle and smear the pitch into a wash. Short SIZE and a sharp SHAPE give a hard, buzzy oscillator; long and soft gives a formant pad. |
Full version. REPEAT is visible only in SYNC.

REPEAT locks each grain to the exact gap between two grid ticks, so the grains butt up against each other instead of overlapping. The result is a verbatim beat-repeat: you hear the sample as itself, chopped on the grid, rather than as a granular texture. It stays locked to the host tempo, so changing tempo re-locks rather than drifting.
While REPEAT is on, SIZE and SHAPE go inert. The grid sets the length and the window becomes a flat top with a short fade at each edge, so neither dial has anything left to decide. They dim rather than disappear, and their tooltips say why.
The ENVELOPE card is a curve with three handles. Drag the first sideways for attack, the middle one sideways for decay and up and down for sustain, and the last one sideways for release. The four values are printed under the curve as you go. Attack runs to 2 seconds, decay to 2 seconds and release to 5 seconds. The handles cannot cross each other.
The line between the handles bends too. Press anywhere on the attack, decay or release segment, away from a handle, and drag up or down: the segment bows toward the top or the bottom of its travel instead of running straight. Dragging up always bows the line upward, on all three segments. Double-click a bend to straighten it back to a dead straight line. The sustain shelf is flat by definition and holds no curve.
This is what turns a linear decay into a snap that falls fast and then hangs, or a release that holds and then drops away at the end. The curves are free, like the rest of ENV 1.
Full version. ENV 1 is free, curves included. ENV 2 and ENV 3 are part of the modulation system.
The tab row on the card's title line reads ENV 1 | ENV 2 | ENV 3. ENV 1 is the amp envelope described above: it is the volume shape of every note, and it is always in circuit. ENV 2 and ENV 3 are not connected to volume at all. They are two more envelopes with the same four handles, the same ranges and the same slope curves, that do nothing until you point them at something.
Assign them from the ROUTING drawer, by picking Env 2 or
Env 3 as a row's source. They have no badge in the footer.
Both are gated by the keyboard rather than by a single voice: every new note retriggers them from the start, and the release only begins when you let go of the last held key. So on a held chord, adding a note restarts the attack for both of them, and lifting one note of three changes nothing. They run from 0 up rather than in both directions, which makes them useful for anything you want opened and then closed once per phrase: a filter cutoff that blooms, a MORPH amount that arrives late, a pitch that settles.
The card always opens on ENV 1. Which tab you were last on is not remembered, and it is not stored in a preset.
Full version.

With one sound already loaded, drop a second file on the lower third of the display. MORPH comes alive, and turning it up starts taking grains from the second sound. Leave the MATCH pill at the end of that lane lit and Murmur transposes the second sound onto the first one's key, which is usually the difference between a morph and a clash. The combo above the MORPH dial decides how each grain chooses between the two:
| Mode | What it does |
|---|---|
| BLEND | Mixes both inside every grain, so the dial is a plain crossfade. |
| SCATTER | Gives every grain a coin flip. |
| ALTERNATE | Hands grains out in strict turn. |
| ONSET | Sends a grain to the second layer when it lands near one of that layer's hits. |
| RESONATOR | Uses the second sound to strike a resonant body tuned to the note you are playing: the morph layer is the hammer, the key is the bell. MORPH fades from dry layer A into the ringing. It wants long grains — see below. |
| FOLLOW | Lets the second sound's loudness open and close the first: layer A sounds where layer B is loud and falls away where it is quiet. MORPH is how hard the gate bites, not a crossfade — see below. |
SCATTER, ALTERNATE and ONSET hand each grain to one layer or the other. BLEND, RESONATOR and FOLLOW instead make every grain out of both at once — which has consequences beyond the sound, two headings down.
The resonant body lives inside the grain, and it can only ring for as long as the grain does. That is the one thing worth knowing about this mode, because at the default SIZE of 50 ms the body is cut off long before it has finished sounding: it measures 13.6 dB quieter than dry layer A, and what you hear is thin enough to mistake for the mode doing nothing. Turning MORPH up does not fix it. SIZE does.
| SIZE | What RESONATOR does there |
|---|---|
| 50 ms | 13.6 dB below dry. Thin, and easily mistaken for nothing. This is the default, so it is where you will first hear the mode. |
| 150 ms | Level-competitive with dry. Treat this as the floor for the mode rather than a suggestion. |
| 400 ms | 10.8 dB above dry, and still sustaining where dry granular has thinned out. This is the setting the mode exists for. |
So reach for SIZE first and MORPH second. Long grains are the point of RESONATOR, not a workaround for it: past about 150 ms it stops being a quiet effect and starts being a sustaining instrument that the second sound is playing.
The body is tuned to the note you play, not to a dial, so it follows the keyboard, TUNE, the pitch bend wheel and GLIDE — which is what makes it playable rather than one fixed colour. Its Q is deliberately not a control: a physical body has a Q, and MORPH already owns the question of how much of the ringing you hear. One consequence worth expecting is that high notes ring shorter than low ones, exactly as a smaller body does.
What you load into layer B is the character of the strike. A broadband, transient-rich sound excites the body across its range and reads as a mallet; a tonal one sitting on the note you are playing drives it hard and reads as a bowed or blown tone. Leaving the MATCH pill lit is what puts a tonal layer B on that note in the first place.
FOLLOW never plays the second sound at all. Each grain runs an envelope follower over the layer B it reads, and that envelope opens and closes layer A: silence where B is silent, wide open where B is loud. Load drums into layer B under a pad and the drums chop the pad in their own rhythm — the morph layer is the player here, not a sound. MORPH sets how deep the gate bites, and the dial covers a lot of ground: the bottom quarter is a gentle pump, the midpoint is already a heavy 30 dB duck, and from there the mode turns into a true gate — at full, the stretches where layer B is silent are silent, not just quieter. Somewhere around 40–50% is the musical middle; past it you are shaping holes, not levels.
The gate's attack is a fixed 2 ms, so layer B's transients bite as transients. Its release starts at 80 ms and tightens as MORPH rises, down to 25 ms at full — low settings breathe between hits, full knob snaps shut so the gaps actually open up. Both are fixed in real time, so a 20 ms grain and a 400 ms grain hear the same gate; short grains simply live less of the release. A quiet layer B means a mostly closed gate rather than a boosted one, which is the honest answer: the gate can only shape the main sound down, never up.
SCAN on layer B picks which part of that rhythm does the gating — scan to a different bar of the loop and the pad is chopped by that bar instead.

In BLEND, RESONATOR and FOLLOW one grain reads both sounds through one window and one gain. It is not a layer A grain and not a layer B grain; it is a single grain made of both. Those three behave as one group everywhere the instrument has to ask which layer a grain belongs to, and every per-layer control stands down in all three of them:
Each one goes flat and inert rather than vanishing, its tooltip names the reason, and your setting is kept. These are modes you pass through: pick SCATTER, ALTERNATE or ONSET and all four come back exactly as you left them.
The other three split dials are unaffected. B PITCH, B POS and B SCAN work in all six modes, because a grain reading both layers still reads them at two rates, from two places, moving two ways. That is true even in FOLLOW, where layer B is never heard directly: B POS picks which part of the second sound's rhythm is doing the gating, and B PITCH speeds or slows that rhythm, because pitch is a read rate.
A morph layer is a second sound, not a copy of the first, and it usually wants its own settings. Four dials — PITCH, POS, SIZE and SCAN — carry a small button underneath that decides which layer the dial is holding. Click it to cycle:
| State | What the dial does |
|---|---|
| AB | Linked, and the default. The dial drives both layers together, which is exactly what it did before there was a choice, so every preset made until now sits here and sounds the same. |
| A | Split. The dial is the main sound's value and moves nothing else. Layer B keeps whatever it was given and does not follow. |
| B | Split, with the dial now holding layer B's own value. The main sound does not move, and the dial's caption changes to say whose value you are turning: B PITCH, B POS, B SIZE, B SCAN. |
The button is only drawn once a morph layer is loaded, and only in the full version: with one sound there is no A and no B to tell apart, so a control that split them would be describing an instrument that is not there. It is quiet at AB, amber at A, and layer B's violet at B — the same violet B's lane, its ticks and its grains already wear.
A and B are one state to the engine: the pair is split, and the button only decides which half you are looking at. Stepping through AB and out again does not wipe either side's value, so you can compare the two layers by clicking back and forth. Both values are host parameters, and both are saved.
B PITCH is layer B's own pitch, not an interval away from A's. That matters because layer B still tracks the keyboard, the pitch bend wheel and GLIDE exactly as layer A does: those three are properties of the note, not of a layer, so they are shared and only the two dials are independent. Scale lock applies to both for the same reason. Drop the same file into both layers, set B PITCH twelve semitones down, and MORPH becomes a fade between a sample and itself an octave lower.
B SIZE cannot apply in BLEND, RESONATOR or FOLLOW. A grain in any of those three reads both layers through one window, so there is no second grain length for it to have. In those modes the SIZE button goes inert, shows AB and its tooltip says why; pick SCATTER, ALTERNATE or ONSET to split it. Your B size is not lost while you are there — the button is locked to AB rather than reset, and splitting again returns the value you had. The other three dials have no such restriction: B PITCH genuinely works in all six modes, because a grain reading both layers still reads them at separate rates inside the one window.
B PITCH, B SIZE and B SCAN are dials rather than modulation destinations, so in state B the
dial carries no modulation rings and a dropped badge is refused. B POS is the exception: it is a
destination in its own right, listed as B Pos in the routing drawer.
The POS button is also the answer to "where does the second sound read". At AB layer B mirrors layer A: wherever the read point sits in sound A, proportionally, is where it sits in sound B, and POS, SCAN and dragging the display move both. Split, B has a read position of its own, and a PIN chip appears beside the button with the third state:
The violet line across the display is layer B's playhead. Drag it to move B along its own timeline without touching where A is reading. Grabbing the line while the two are still linked splits the pair and starts the drag in one motion, so you do not have to press the button first. Switching PIN never jumps the line: the number behind it is rewritten so B stays where it is while its meaning changes from a distance to a place. Going back to AB snaps B onto A and keeps the distance you had set, so splitting again returns B to where you left it rather than making you find the spot twice.
PIN holds B's base position; JITTER still scatters each grain around it. And in SLICE or WALK mode, once layer B has slices of its own, PIN changes nothing — each key already addresses B's own slice at its own absolute place, so the two layers are independent there whatever the chip says. For the same reason the POS dial is inert on the B side in those modes: see section 6.
The grain cloud follows: B's grains are drawn at B's position, not A's, so you can see the two layers reading different parts of two different files.
A second sound is cut up by its own transients, not the first one's. With a morph layer
loaded there is a small link pill between the SLICES dial and the space to its right, reading
A ↔ B while the two are joined. Click it and a second SLICES dial appears
there, with small A and B captions under the pair, and from
then on layer A's sensitivity divides A and layer B's divides B — six slices on one and
five on the other, if that is what the material wants. Linked, the one dial divides both and the
second is not drawn, because a dial showing a value that is not in force would be a lie.
What each layer then does with those slices is in section 6.

One DENSITY dial feeds both sounds, which is the right answer until you want
a sparse second sound punctuating a dense first one rather than two clouds of the same thickness.
The chip under DENSITY cycles 1:1, 1:2, 1:3 and
1:4, and it decides how often the morph layer actually takes the grains it is
handed.
Layer A is untouched. The grains layer B does not take are dropped, not handed back, so thinning B never makes A denser and never moves a single layer A grain. The main sound goes on exactly as it was while the second one gets sparser, which is what lets you set the balance by ear without the first sound shifting under you.
The count is of the grains nominated for layer B, not of every grain in the cloud, so the ratio means the same thing whichever splitting mode you are in. In ALTERNATE with MORPH at halfway, where layer B already has every other grain, 1:2 gives it every fourth grain overall — rather than all of them or none, which is what counting the whole stream would have done.
The chip is drawn once a morph layer is loaded, and it is one of the controls that stands down in BLEND, RESONATOR and FOLLOW, for the reason given earlier in this section: no grain belongs to layer B there, so there is no second rate to set. It keeps showing the ratio you picked while it is inert and hands it straight back when you choose a mode that can honour it. It is a host parameter like the rest, saved with your session and with your presets.
Murmur analyses every sample you load and marks the hits it finds. Those marks are the amber ticks on the waveform, and one set of them does three jobs: they are the slices SLICE mode plays, the targets MAGNET pulls toward, and the steps WALK walks through.
The GRID combo under the SLICES dial decides where they come from. It reads TRANSIENT — the hits Murmur heard — or a musical division, which puts them on the beat instead. Either way you can move, add and remove boundaries by hand afterwards.
At TRANSIENT the ticks are wherever Murmur heard a hit, and the SLICES dial sets how fussy that is. Low keeps only the strongest transients, high keeps everything. This is the mode you want on a drum loop with clear hits.
Three things about what the detector hears, because they decide where the ticks land. A hit that starts at the very beginning of a file is detected: on a loop cut on the downbeat, the first hit is a real slice, and C1 in SLICE mode plays it rather than the second hit. A file that fades in, or a drone that simply starts, is not treated as a hit at sample 0 — steady material has no onsets, which has always been the rule. How strong a hit counts as is measured from the hit itself, from the energy it adds, rather than from how far it stands out against whatever came immediately before it. That is what makes the SLICES dial ramp: hits of the same weight now rank the same whether they land in silence or on the tail of the one before, so a dense passage thins out gradually instead of vanishing in one click of the knob. And there is a ceiling on how many transients Murmur will hold for a file — when a file has more than that, the strongest are kept, not the first, so a long recording no longer runs out of slices partway through and leaves its last stretch bare.
When Murmur is asked for a set number of boundaries rather than for everything above a threshold — RE-SLICE, or a STEPS count in the WALK drawer — it now also keeps them spread out. It still takes the strongest hits first, but a hit has to sit at least half an average slice width away from every boundary already placed before it is accepted. Without that rule one loud bar took every slot and the rest of the file got nothing. If the rule is too strict to fill the count, Murmur relaxes it and tries again rather than handing back fewer boundaries; it still splits the widest remaining gaps if it comes up short, and it still falls back to an exact even division on material with no hits worth finding at all, which is why this works on pads and drones. The trade is that on genuinely dense material a real transient in a busy passage can be passed over in favour of a weaker one in a quiet stretch.
Pick a division in GRID instead and the boundaries stop being about the
audio at all. Murmur lays them at exact multiples of that division from the start of the file,
at your host's tempo, all the way to the end. The list runs 1/1 down to
1/64, each with a dotted (1/8.) and a triplet (1/8T)
variant. With no host clock — the standalone, or a host that reports no tempo — it uses 120 BPM,
exactly as the SYNC grain clock does.
This is how you chop a bar into sixteenths whatever the sample does, and it is the only way to get even boundaries on material where detection has nothing to hear. The GRID combo applies to the first sound only; layer B's boundaries are reached through its own CLEAR / DETECT pill and by editing its lane.
There is a ceiling of 256 boundaries, and a division covers 256 spacings of the file and no further. If the division you asked for would run out before the end of the sample, Murmur uses the finest division that does cover the whole file and moves the combo to show it, so what is written on the control is always what is in force. At 120 BPM that puts the limit at about 32 seconds for 1/16 and 8 seconds for 1/64: on ordinary loops you will never see it, and on a long file you will see the combo settle on something coarser than you clicked.
In SLICE or WALK mode, double-click the waveform to add a boundary where you clicked, or double-click an existing tick to take it away. Zoom in first if the ticks are close together, because the grab radius is measured on screen rather than in the file. The limit is 256 boundaries, and Murmur says so over the display when you reach it.
Once the boundaries are authored rather than detected — after a GRID division, a STEPS count or a CLEAR — every boundary is draggable. Press within a few pixels of a tick and drag it along the waveform; it stops before it can pass its neighbours. Even boundaries are a starting point you move, not a decree. SLICES goes dim while the boundaries are authored, because the count is no longer its decision, but it still guides RE-SLICE.
At the right of each lane's own row is a small pill that names where that lane's boundaries come from, and switches it. Layer A's rides the top row and appears in SLICE and WALK mode, the modes where its slices do something; layer B's rides the B lane header whenever a morph layer is loaded.
The pill is lit in the lane's own colour while the set is yours, because a lane whose slice count no longer answers to the SLICES dial should not be a quiet fact.
The STEPS combo in the WALK drawer takes a number from 1 to 16 and asks for exactly that many boundaries. A count you authored wins over the GRID setting: GRID decides where the boundaries are placed, STEPS decides how many there are, and the boundaries stop when the count runs out rather than continuing to the end of the file. Setting STEPS back to AUTO drops the pattern and the authored boundaries together and puts detection back in charge, whatever the GRID combo happens to read. Move GRID to lay a fresh set down.
Two buttons in the WALK drawer, deliberately separate:
In SLICE mode the keyboard maps onto the boundaries from C1 upward, one slice per semitone, each at its natural pitch. POS is dimmed because the key you press is now the position, and dragging the display no longer scrubs. Notes below C1, or past the last slice, are silent rather than wrapping around. Play the loop back in a different order, or hold three keys at once.
With a morph layer loaded, SLICE and WALK do not point layer B at a proportion of where layer A is reading. Each layer reads from its own detected transients, addressed by the same number. Key 3 plays layer A's third slice and layer B's third slice, wherever those two happen to fall in their two files; in WALK it is the step's slice number that both layers look up. Layer B has its own SLICES sensitivity (section 5), so the two can have different counts, and the ordinal is the only thing they share. That is the point: a five-hit vocal chop and an eight-hit drum loop stay two instruments you play from one keyboard rather than one instrument dragging the other around.
It follows that a key can run off the end of one layer and not the other. A key past a layer's last slice is silent on that layer rather than wrapping around or falling back to the other one's slice. In BLEND and RESONATOR that grain simply loses its B share and leaves dry layer A where it was; in FOLLOW a layer B with nothing to play is a gate that never opens, so layer A comes through held at the depth floor — quiet at low MORPH, silent at full; in SCATTER, ALTERNATE and ONSET the grains dealt to B are dropped. Layer A keeps playing normally, so the top of the keyboard thins out rather than breaking.
And the POS dial is inert for both layers here, not just for the main sound:
the slice is the anchor, so a dial that moved the read point would be dragging a layer off the
slice it was told to play. Modulation still applies, and adds around the anchor rather
than replacing it — a route to Position moves layer A around its slice and a route
to B Pos moves layer B around its own. A dimmed dial and a live route are the same
rule on both layers.
Full version. In the free version the WALK segment is dim and the mode falls back to CLASSIC.
In WALK mode the read point steps through the slices on its own, in time with the host. You still have to hold a note for anything to sound: WALK moves the read head, it does not play notes. It ignores which key you hold, so any note gives you the same phrase at the sample's own pitch, unless TUNE is on.

Switch to WALK and a fourth tab, WALK, appears at the bottom right. It opens a short drawer that deliberately stops below the waveform, so you can see the boundaries while you edit the pattern that points at them.

Out of the box STEPS reads AUTO, which means there is no pattern: WALK simply
walks whatever detection found, in order. The row still shows you what it is doing, dimmed to say
it is a readout rather than something to edit. If detection found more than sixteen slices, a
+N tag on the right stands in for the ones off the end and lights up when the walk
is out among them.
Clicking any cell at AUTO takes control: it authors a pattern with that many steps and hands you the editable row. Picking a number from STEPS does the same thing deliberately.

Pick a number in STEPS. Murmur lays out that many boundaries on the waveform and that many cells in the row, each pointing at its own slice: step 1 plays slice 1, step 2 plays slice 2, and so on. That is your starting point. Now change it:
Select a step and four fields appear underneath it. They apply to that step alone.
| Field | Range | What it does |
|---|---|---|
| OFFSET | 0 to 100% | Nudges the start point inside the step's own slice, as a fraction of the distance to the next boundary. It cannot leave the slice, and it means the same musical amount whether the slices are 60 ms or two seconds apart. Use it to push a step slightly late without moving the boundary itself. |
| LEVEL | 0 to 100% | Volume for this step. Ghost notes. |
| PITCH | ±24 st | Transposes this step. Set in cents, so you can detune slightly as well as jump an octave. Scale lock still applies on top. |
| PROB | 0 to 100% | The chance this step sounds on any given pass. The dice are seeded, so a bounce renders the same twice and the same bar sounds the same every time round. |
Each field is drawn on the cell as well, so you can read a pattern at a glance without selecting anything: the bar along the bottom is LEVEL, the tick above it is OFFSET, the chevron in the corner is PITCH, and the small ring is PROB. Marks only appear when the value is away from its default. Double-click a field to clear it.

ORDER is Forward, Ping-Pong or Random, and it reorders the steps, not the slices, so it works on the pattern you built rather than undoing it. Ping-Pong bounces without repeating the end points. Random is seeded from a fixed number, so it scrambles but it repeats: the same bar plays the same way every time, and a bounce matches what you heard.
RATE is 1/1, 1/2, 1/4, 1/8 or 1/16, and defaults to 1/8. WALK locks to host tempo and to the transport position, so steps land on the bar. With no host clock it free runs at 120 BPM. Note that it keeps stepping when the transport is stopped.
Full version, and only with a morph layer loaded.
With a second sound in the violet lane, a pill appears in the control row between STEPS and the buttons. It has two states, and it decides whether the morph layer is following the pattern above it or walking one of its own.

The violet row edits exactly like the amber one: click a cell to hear it and hold it, drag sideways to point it at a different slice, double-click to switch it off, and the four fields underneath belong to whichever step you last selected. Only one step is selected at a time across the two rows, and selecting one on either row parks both layers — each on the step that position names in its own row, which is how you hear the two against each other while you edit.

Both rows run on the same step clock. There is deliberately no second RATE, and no second ORDER either: those two describe the shape of the walk, and the whole point of the split is that the two rows are the same walk over different material. The drift comes from the lengths. Four steps against six come back round every twelve; five against sixteen take eighty. Set B STEPS to something that does not divide into STEPS and you have a polyrhythm you did not have to program.
You get one for free before you author anything. Leave B STEPS on AUTO after splitting and layer B walks its own detected slices, in order, exactly as layer A does at AUTO — and the two sounds rarely detect the same number of hits, so they are already out of phase with each other.
Layer B's steps address layer B's own slices, by number. The two sounds can be sliced completely differently — a different count, a different sensitivity, boundaries you dragged yourself — and B's step 3 still means the third slice of the second sound. The numbers wrap into B's own count the same way A's wrap into A's.
RESET clears both patterns, because it has one meaning and a button that quietly reset only the row you last touched would be worse than one that resets both. RE-SLICE re-detects the boundaries on the main sound, as it always has.

In MORPH's BLEND, RESONATOR and FOLLOW modes a grain is not one layer or the other: every grain reads both sounds through one window and one gain. There is no layer B grain for a layer B step to be about, so the pill goes dim and inert, the violet row disappears, and the engine runs the single pattern. Your choice is not thrown away — switch to SCATTER, ALTERNATE or ONSET and the split comes back exactly as you left it, steps and all. Section 5 covers what those three modes have in common and which other controls stand down with them.
For the same reason the split needs grains actually reaching layer B: with MORPH at zero every grain is layer A's, and B's row is running but silent. Turn MORPH up to hear it.
Both patterns and the pill's state are saved in your DAW session and in any preset you save, and the pill is a host parameter you can automate.
The pattern and the boundaries are stored separately on purpose. The boundaries travel with the sample; the pattern travels with the parameters. That means dragging a boundary does not disturb the pattern, and a step that pointed at slice 3 still points at slice 3, which is now somewhere else. It also means you can point more steps at fewer slices: the numbers wrap rather than clamp, and shrinking the slice count and putting it back restores exactly what you had.
Both are saved in your DAW session and in any preset you save.
Full version. All of it: the sources, the rings and the matrix.

The MOTION card holds four pages, and the tabs for them sit on its title line: LFO 1, LFO 2, FILTER and COMP. The first two are the LFOs described below. The last two are the per-grain filter and compressor, which are not modulation sources at all and have a section of their own (section 9); they live here because there was room and because their cutoff is the thing you most often want an LFO on.
The four macros and RANDOM are pinned under the pager and stay put on all four pages, so turning a macro never means going back to a tab first.
The page you are on is a place to stand, not a sound. It is remembered with your session, and loading a preset deliberately leaves it alone rather than moving the card out from under you.
There are ten sources. Eight of them live in the footer as draggable badges; ENV 2 and ENV 3 are assigned from the ROUTING drawer instead.
8 GR. It is the way to get "every fourth grain is
darker" rather than "the cloud gets darker twice a second".The rate dial follows the clock, so it is never
showing a number that is not in force. In free mode it sets the rate in hertz as it always
has. In sync it steps through the musical divisions and reads one of them, 1/8.
In grain sync it steps the grain counts and reads how many grains a cycle takes. Its tooltip
changes with it, and right-clicking the dial opens the same clock menu the
trace does, which is the short way to change clock without going to the trace first. An
LFO's own rate is the instrument's clock rather than something a route drives, so the dial
is not a modulation destination in any of the three.
The five shapes are not the only shapes. There is a DRAW chip in the top left corner of each LFO trace, and clicking it turns the trace into an editor for that LFO's own curve. It opens on a starting shape rather than a blank strip, so there is always something to pull at.
The clock menu still works while you are drawing: right-click the background, or click the tag in the corner. And the five standard shapes stay one click away underneath — picking one leaves the drawn curve behind and goes back to that waveform. Your points are kept, so DRAW brings the curve back exactly as you left it.
Drag a badge onto a knob to make that source move it, and keep dragging up and down before you let go to set the depth. After that the route is a coloured ring around the knob: drag the ring to change depth, double-click it to reset, right-click it to remove. Right-click the middle of a knob for a menu of everything reaching it. Macros are sources, so dropping a macro badge on another macro does nothing and Murmur says so.

The matrix gives you source, destination, depth from -1 to +1, a layer button covered below, and an x to clear the row. It is the only place to reach a destination that has no knob on the front panel, and the only place to assign ENV 2 and ENV 3. ONSET at the bottom sets how grain starts are timed when the clock is HZ: Periodic spaces them evenly, Poisson randomises the gaps for a cloudier feel.
There are sixty-four route slots and the drawer shows sixteen of them at a time, as two columns of eight, so the table is paged. Four numbered buttons on the drawer's title line, over on the right beside CLOSE, move between the pages: page one is slots 1 to 16, page two is 17 to 32, and so on. The slot numbers down the left of each column are absolute, so the row labelled 20 is the twentieth route wherever you are standing. Hovering a page button says which slots it holds and how many of them are free.
Beside those buttons a count reads n / 64 USED. It is the whole table's count rather than the current page's, because the question it exists to answer is whether there is room left anywhere. A page with no free slots dims its own number, which is how you find the space without opening all four.
You will not often go looking, because dragging a source badge onto a knob fills the first empty slot wherever it is, on whatever page that turns out to be. The pages are for reading and editing the table, not for feeding it. Which page you are on is simply where you are standing: it is not a host parameter, it changes nothing about the sound, and a freshly opened editor starts on page one.
The destination list is grouped, because forty-odd names in one flat column is a scan rather than a choice. The plain grain controls come first, ungrouped. Then two headings for the per-grain processors in section 9 — GRAIN FILTER with Cutoff and Res under it, and GRAIN COMPRESSOR with Across and Within — because on the panel those dials live behind FILTER and COMP tabs and were the hardest four to find in a flat list. Then the FX rack, a module at a time under its own heading, in the order the signal takes. The grain compressor's two dials are new to the list: they were the last continuous per-grain controls with no destination at all, and a per-grain compressor whose amount cannot change per grain was missing the point of being per-grain.
Three destinations exist only here. They have no knob on the panel, they sit at a neutral value and stay there, and a route is the only thing that ever moves them. That is not an oversight: each of them is per-grain by nature, and a single dial for a value that wants to be different on every grain would be the wrong control.
| Destination | Range | What it does |
|---|---|---|
| Pan | -1 to +1 | Where in the stereo field a grain sits, before SPREAD scatters it. It rests dead centre. SPREAD then throws each grain left and right around wherever Pan is, so the two stack: with SPREAD at zero, a route on Pan moves the whole cloud as one, and with SPREAD up, the scatter rides on top of it. It is worth knowing that a random source on Pan is very nearly what SPREAD already does — the routes that earn their place are the slow and the deliberate ones. An LFO drifts the cloud side to side. Keytrack puts low notes to the left. And since Pan takes a layer scope, you can sit the main sound left and the morph layer right. |
| Level Rnd | 0 to 36 dB | How much each grain may be randomly quietened by. Every grain draws its own amount between nothing and this, so it only ever takes level away. A little is a flicker over the cloud that keeps a steady stream from sounding mechanical; a lot drops grains almost out and thins the texture into glitter. The SHIMMER macro is wired to it out of the box, which is where that macro's sparkle comes from. |
| Level | 0 to 2 | A plain gain on each grain, resting at 1. Use it when you want a source to shape loudness per grain without the randomness Level Rnd brings. |
Full version.
Every row in the routing drawer carries the same AB / A / B button the four dials in section 5 wear, and it answers the same question: which layer this route is allowed to touch. So LFO 1 can modulate the main sound while LFO 2 modulates the morph layer, from one instrument, with one keyboard. AB is the default and what every preset made before this loads as, so nothing you already have moves.
Unlike the dials' buttons, this one stays clickable with no morph layer loaded, because a route can be set up before the second sound arrives. It has nothing to act on until then: with one sound every grain is a layer-A grain, so a route set to B is silent until you load a layer for it to reach.
A scope bites where the engine decides a value after it knows which layer the grain is reading. Some destinations are settled before that point or are not per-grain at all, and those cannot honour a split — their button is drawn flat and unclickable on AB rather than offering a choice that does nothing, and its tooltip names the destination and says the route reaches both layers.
| Group | Destinations |
|---|---|
| Takes a layer | Pitch, Spread, Shape, Pan, Level Rnd, Level, and all four per-grain processor controls: Grain Cutoff, Grain Res, Grain Comp Across, Grain Comp Within. |
| Always both | Density, which drives the one grain clock both layers are spawned from, and the whole FX rack, which runs once per block on the summed output long after any grain has a layer. Plus the ones that place the grain in the first place — Position, Grain Size, Jitter, Magnet, Cohere, Scan, Morph and B Pos — because in SCATTER and ONSET those are what decide which layer the grain reads, so scoping them would be circular. |
BLEND, RESONATOR and FOLLOW ignore scope entirely. A grain in any of those three reads both layers through one window, so it is neither A nor B and every route reaches it. In those morph modes the button goes inert on every row and says why. Your choice is not thrown away while you are there — they are modes you pass through, and the row still shows what the preset actually carries — it simply has nothing to act on until you pick SCATTER, ALTERNATE or ONSET.
One shortcut: a route you make by dragging a badge straight onto one of the four split dials starts on the layer that dial's own button is showing, rather than on both. In practice that only shows itself on PITCH — drop a source on it while it reads A and the route is scoped to A. Position, Grain Size and Scan are on the "always both" list above, so those three dials make an AB route whatever their button says, and in state B they are not modulation targets at all.
Full version. Both of them.
The FILTER and COMP pages of the MOTION card hold two processors that are not part of the FX rack and do not work like it. The rack in section 10 sits after everything, on the finished stereo mix, once per block. These two run inside the engine, on each grain as it is generated, with that grain's own state.
That difference is the whole point of them. A filter on the mix has one cutoff at a time. A filter on the grain has a cutoff per grain, so it can be a different value on every single one — which is why the headline route for it is GRAIN RND onto Grain Cutoff, and why a grain compressor can even out grains taken from a sample whose level wanders, something no compressor on the output can do.
| Control | Range | What it does |
|---|---|---|
| MODE | 5 types | Off, Low Pass, High Pass, Band Pass or Notch. Off is a true bypass: the filter is not in the grain's path at all. |
| Cutoff | 20 Hz to 20 kHz | Where the filter bites. Each grain takes its own copy of this value at the moment it is born and keeps it for its whole life, which is what makes it worth modulating per grain. |
| Res | 0.5 to 8 | How sharply it rings at the cutoff. 0.7 is flat; up high it whistles at the cutoff pitch. |
The filter runs on the raw grain before its window is applied, so the window still takes the grain cleanly to silence at both ends and nothing rings past it into a click. Each grain starts with the filter cleared, so it can never inherit the ringing of the grain before it.
The two dials are also modulation destinations, named Grain Cutoff and Grain Res in the matrix. Cutoff is modulated in octaves rather than in hertz, so a route at full depth sweeps the whole range and a small depth means the same musical amount wherever the dial is parked. Drop the GRAIN RND badge on the cutoff dial and every grain lands somewhere different: at a low depth this is a shimmer over the cloud, at full depth every grain is a different colour.
The dials go dim while MODE reads Off. That is the only reason they dim other than the licence.
Two knobs and a mode. The knobs are two different jobs, not two amounts of the same job, and both are off at zero.
| Control | Range | What it does |
|---|---|---|
| ACROSS | 0 to 1 | Evens the loudness between grains. Each grain gets one gain, decided when it is born, from how loud the sample is where that grain is reading. Grains lifted out of a quiet passage arrive at the same loudness as grains from a loud one, so scanning through a sample whose level wanders stops being a ride on the fader. |
| WITHIN | 0 to 1 | Compresses the dynamics inside a single grain, flattening its own attack and decay as it plays. Long grains hold instead of falling away; a grain cut across a transient stops being mostly transient. |
| MODE | 3 types | Shared by both knobs, so it means the same thing on either. See below. |
| Mode | What it does |
|---|---|
| Radical | Flattens hard, and starts doing it in the first third of the knob. All the way up, every grain lands on the same level. |
| Threshold | A traditional compressor. Grains above the threshold are pushed back down, quiet ones are left alone, and an automatic make-up gain brings the whole thing back up so a normal grain comes out where it went in. |
| Inverse | The same curve pushed past flat, so the order comes out backwards: loud grains end up quieter than quiet ones. Past halfway on the knob the sample's own dynamics are turned inside out. It is a sound design effect, not a gentler setting. |
Everything is limited to 24 dB in each direction, so a near-silent grain is not lifted into the noise floor and a loud one is not annihilated. WITHIN uses a fast attack and a short release fixed in real time rather than in grain lengths, so a 20 ms grain and a 400 ms grain hear the same compressor; the short one simply does not live long enough to get far into the release.
Full version.

Every strip is built on the same grid, whether the module has four controls or five, and the rows start at the same height in all seven — so the rack lines up across as well as down, and the bottom row is always that module's MIX, read as its output stage.
The dot in each header switches that module off and back on. A module sitting at its neutral value is skipped entirely, so an unused rack costs nothing.
BYPASS at the top right of the drawer, next to CLOSE, takes the whole rack out in one control without touching a single setting. It is not a mix and not a second fade: it is the same skipped chain you would get by pulling every module's mix to zero, so the rack costs nothing while it is bypassed and everything is exactly where you left it when you switch back. It is a host parameter like the rest, so you can automate it.
The small AB chip in each module's header, between the name and the dot, points that module at one layer: AB processes the whole instrument, A processes the main sound and B processes the morph layer. It is the same chip, and the same amber-for-A, violet-for-B colour, that the dials and the routing rows wear. The layer a module is not pointed at simply goes round it, so it gets no quieter, only less processed — reverb on the pad and none on the sample, a filter that eats one sound and leaves the other alone, and both at once in the same chain.
The module ORDER is still one chain shared by both layers; what changes is which modules each layer passes through. Point every module the same way and the rack runs once, exactly as it always did. Point them different ways and the chain runs twice, once per layer, each with its own filters, delay lines and reverb tails — so a module on AB gives each layer its own tail rather than pouring one into the other. That second pass is the only thing here that costs anything, and you only pay for it when you split.
Changing a chip takes the same short fade an effect switching on already takes, and tails carry on ringing. Two things it cannot split: a grain in MORPH's BLEND, RESONATOR or FOLLOW mode reads both layers through one window and cannot be cut in half, so those grains stay in the rack whichever layer you point a module at, and with no morph layer loaded there is only one sound to process, so every chip sits flat at AB until you load one. They are host parameters, saved with the preset.
| Module | What it does |
|---|---|
| FILTER | One knob whose centre is always off, and a TYPE that decides what the travel either side of centre does. At LP/HP, left darkens with a low pass and right thins with a high pass. At Band, both directions sweep the same band pass, from 20 Hz at the centre up to 20 kHz at either end, and RES becomes its bandwidth. RES adds resonance, SLOPE picks 12 or 24 dB per octave, MIX blends the filtered signal back against the untouched one. This is the filter on the finished mix; the one in section 9 is a filter per grain. |
| SATURATE | Pushes the cloud into soft distortion, which adds warmth and grit. DRIVE is how hard it is pushed, TONE tilts that dark or bright, MODE is Soft (rounded, tape-like) or Hard (clipped), MIX is parallel so you can drive it far harder than you would want to hear alone. |
| SQUASH | Holds the loud parts down and lifts the quiet ones up, which turns a sparse cloud into a wall. SPEED sets how fast it reacts, TILT leans it toward one job or the other, MIX gives you parallel compression. |
| CHORUS | A slowly detuned copy alongside the cloud, so it widens instead of sitting still. RATE, DEPTH, FB and MIX. |
| PHASER | Moving notches sweeping through the cloud. RATE, DEPTH, FB and MIX. |
| DELAY | Tempo-locked echoes. TIME is a division from 1/16 to 1/2 including dotted 1/8, MODE is Normal, Ping-Pong, Reverse or Glitch, then feedback and mix. |
| REVERB | Small room up to endless tail. PRE, SIZE, DAMP and MIX, plus SHIMMER, which feeds the tail back through an octave-up shifter so it keeps climbing instead of fading. PRE is pre-delay, 0 to 200 ms: how long the tail waits before it starts, which is what keeps a sound in front of its own space rather than buried in it. At 0 ms the delay line is skipped entirely. |
This is the setting that makes Murmur sound like the record rather than like a granular plug-in, and it is easy to miss by ear because nothing on the panel is labelled as it. Load a drum loop and set:
| Control | Value |
|---|---|
| Grain clock | SYNC |
| Division | 1/16 |
| SIZE | 125 ms |
| SHAPE | exactly centre |
| JITTER | 0 |
| SCAN | 0 |
Three things have to be true at once for this to work, which is why it is hard to stumble on. The grains have to arrive on the grid, which is what SYNC and 1/16 do. Each grain has to be exactly as long as the gap between them, so that they butt together and neither overlap nor leave a hole: at 120 BPM a sixteenth is 125 milliseconds, which is where that number comes from. And the window has to be flat, which is what SHAPE at centre gives you. The centre of the SHAPE dial is a flat top window that passes the middle of each grain completely untouched and only tapers the last fraction at each edge. Anywhere else on that dial and every grain gets a fade over its whole length, which is what smears a chop into a texture. JITTER and SCAN at zero keep the read point still so the pieces come out in order.
The one weakness is the 125: it is correct at 120 BPM and wrong everywhere else, so at 90 BPM the grains fall short of the grid and you hear gaps. REPEAT is this recipe made automatic. It measures the actual gap between grid ticks at your actual tempo and sets every grain to exactly that, which is why it takes SIZE and SHAPE away from you while it is on. Use the recipe to understand what is happening; use REPEAT when you want it to keep working after you change the tempo.
Turn JITTER up until the grains scatter and the groove falls apart, then bring MAGNET up. The scatter stays, but every grain is dragged onto a real hit, so it lands in time. Somewhere in the middle is a loop that breathes without losing the pocket. Use SLICES to decide how many hits count as targets.
Load something pitched: a vocal note, a sustained synth, a bowed string. Check the root
readout has found a key — COHERE trusts the audio's own pitch analysis, never the filename, so
a sample named Gm.wav only locks to G if the sound actually repeats there. Now
bring COHERE up. Grain timing, position and length lock to the pitch period
and the cloud folds into a solid, oscillator-like tone you can play chords with — at the same
loudness, so what changes is the character, not the level. Back it off and it dissolves into
texture again.
How hard it locks follows how sure the detection is. A clear note gets the full lock, a weakly pitched texture gets a proportionally gentler pull rather than nothing, and on material with no usable pitch the dial dims and its tooltip names the reason. One honest note: on strongly pitched simple tones the middle of the travel can dip and comb a little — that is the partial lock, halfway between scattered and phase-aligned grains — so the sweet spots are the low third, where it is still texture, and the very top, where it is a tone.
Slice a bar into 8 with STEPS, then in the sequencer point two steps at the same slice, switch one step off and set another to 50% probability. Set ORDER to Ping-Pong. You now have a phrase that repeats over four bars instead of one, and because the probability is seeded it still renders identically every time you bounce it.
Set the grain clock to KEY. The grain rate now follows the note you play, so the stream itself is the pitch and DENSITY steps aside. Whatever key the sample is in stops mattering: you are playing the grain train, not the recording, so it is in tune chromatically across the keyboard. Keep SIZE short, somewhere around 20 to 40 ms — at KEY rate the grain rate is the note frequency, so a long grain overlaps many copies of itself per cycle and the pitch smears into a wash. Short SIZE and a sharp SHAPE give a hard, buzzy oscillator; long and soft gives something more like a formant pad. The Tone template is this setting ready made.
On the MOTION card's FILTER page, set MODE to Low Pass and park the cutoff somewhere in the middle. Then drag the GRAIN RND badge from the footer onto the cutoff dial and pull the depth up as you let go. Every grain now takes its own cutoff at the moment it is born and keeps it, so a cloud that was one colour becomes hundreds of them. A small depth is a shimmer over the top; full depth spreads the grains across the whole range. Turn DENSITY down until you can hear individual grains while you set the depth, then put it back.
Click the preset name in the header to open the browser. There is no separate PRESETS button. Click a name in the list to load it. Type a name and press SAVE to save, which asks before overwriting. DELETE removes the selected preset, and INIT resets every parameter, restores the factory macro routes and unloads the sample. The chevrons either side of the name step through the list without opening the browser at all.
The list is not one straight alphabetical run. Presets that carry no audio come first, then the ones with a sample inside them, and each half is alphabetical within itself.
The reason is what a preset without audio does when you load it: nothing to the sample. It keeps whatever you already have and reshapes it. That is the whole point of a template, but it makes the preset a bad neighbour to a preset that does bring its own sound. Step off a drum kit straight onto a settings-only preset written for a pad and you hear the drums through pad settings, which sounds like the preset is broken when the preset is fine. Front-loading them means a first walk down the list meets them against whatever you loaded yourself, where they make sense, and every preset after that brings its own material with it.
The grouping is read from the preset file rather than from its name, so a preset you saved a second ago lands in the right half with nothing to rename. The browser and the chevrons read the same one list, so what you see and what stepping walks cannot disagree.
Nine presets are installed with Murmur: Start Here, Soft Pad, Frozen, Grain Cloud, Chop Kit, Beat Grid, The Chop, Magnet Drums and Tone. They are templates rather than sounds: each one carries settings and no audio at all, so loading one leaves whatever sample you already have in place, and dropping a new file in gives you your own material shaped that way. Each is built from free controls only, so all nine play in full on the free version.
| Template | What it sets up |
|---|---|
| Start Here | The plainest useful sound: mid grains, a little SPREAD, nothing exotic. TUNE on. |
| Soft Pad | Long grains (420 ms), a soft window and a slow envelope, so chords hold and bloom. |
| Frozen | Almost no JITTER and no SCAN: one point in the sample, held still, with a long attack and release. |
| Grain Cloud | Short grains with JITTER and SPREAD up, for a scattered, airy field. |
| Chop Kit | SLICE mode on detected transients, one key per slice, with a fast envelope. |
| Beat Grid | SLICE mode on a GRID division instead of transients, so the chops are even whatever the sample does. |
| The Chop | The chop recipe from the recipes section, ready made: SYNC, SIZE 125 ms, SHAPE dead centre, JITTER and SCAN at zero. |
| Magnet Drums | JITTER on and MAGNET at 0.85, so every grain is pulled onto a transient. Made for loops. |
| Tone | KEY rate with 26 ms grains: the grain train is the note, so you play it chromatically. |
They live in a folder the installer owns and Murmur never writes to
(macOS /Library/Application Support/Sounds Like You/Murmur/Presets/,
Windows C:\ProgramData\Sounds Like You\Murmur\Presets\), and a preset of your own
with the same name takes precedence in the list.
A preset is a .murmur file, and the sample
travels inside it, whatever its length. That means you can
send someone a preset and it will open on their machine with the sound intact, without hunting
for the original file. The audio is stored losslessly (24-bit FLAC), which costs about 5 MB
per minute of sample for typical material at 48 kHz, and about 9 MB per minute at
most. Your DAW session stores the audio the same way, so reopening a project never asks you to
find anything.
Drop a .murmur file someone sent you into your preset folder from section 1 and
it appears in the browser. If it was made with parts of the full version you do not have, it
still opens and keeps everything it holds; those parts simply sit inert until a key is
entered.
Murmur is free, and a key unlocks the rest of it. Section 2 has the list of what a key adds. There is no trial: nothing counts down, nothing expires, and the free version never goes silent.

Click UNLOCK to open the licence panel. Nothing else opens it: Murmur never interrupts you with it.

Your key arrives by email after you buy. It is one long line starting MURMUR.
There are two ways to enter it, and they do the same thing:
On success the panel says so, the UNLOCK pill and BUY disappear, and every unlocked control comes alive immediately. There is no restart, no reopening the plug-in and no reloading the session. If you were part way through a track, keep going: a morph layer or a WALK pattern that was sitting inert in the session starts working on the spot.
If the key is not accepted, the panel stays open and tells you. The most common cause is a
partial copy, so select the whole thing including MURMUR. at the front.
Nothing is sent anywhere. The key carries a signature, and Murmur verifies it against a public key compiled into the plug-in. That is arithmetic on your own machine. There is no activation server, no account, no online check, no machine fingerprint and no limit on the number of your own computers you install it on. The licence is for one person, though, and can't be shared (clause 2.2 of the licence agreement). It works with no internet connection, now and in ten years.
The only time Murmur ever contacts the network is the once-a-day update check described in section 1, which carries no identifier and no licence data.
The one thing that can end a licence is a refund. If you get your money back, the licence goes with it (clause 8.3 of the licence agreement) and Murmur returns to the free version.
That adds nothing to the request. The once-a-day file in section 1 also carries a list of refunded licences, as one-way hashes rather than anybody's key, email or order number, and Murmur compares its own licence against that list on your machine. Your key never leaves the computer, and the website is never told who is asking.
Nothing is deleted and nothing goes quiet. Your presets, samples and sessions are exactly as they were, the free version keeps working the way it always has, and buying again opens the rest back up. If the check cannot reach the website, or what comes back is not a valid list, your licence is left alone: only a clear match in a well-formed list changes anything.
Your key is stored per user, outside your projects:
macOS: ~/Library/Application Support/Sounds Like You/Murmur/settings.properties
Windows: %APPDATA%\Sounds Like You\Murmur\settings.properties
It is deliberately not stored in the plug-in's state, so a preset made on a licensed machine carries no licence, and a session file never contains your key. Every instance of Murmur on the machine reads the same one, so you enter the key once.
Quit the DAW, then rescan plug-ins. Most hosts have a rescan or "clear cache and rescan"
button in their plug-in preferences. Check the file really is in the folder listed in section 1.
On Windows, make sure you copied the whole Murmur.vst3 folder and not just the
file inside it. In Logic and GarageBand the first launch can take a minute while the Audio Unit
is validated.
Check a sample is loaded: the file name shows in the top left of the display. With nothing loaded there is nothing to granulate. Then check OUT is up and the meter is moving. The licence is not a cause: the free version is never silent, so if you are getting nothing at all it is something else.
One case that looks like silence but is not: in SLICE mode the keyboard starts at C1 and runs upward one slice per semitone. Notes below C1, or above the last slice, do not play. Try playing around C1.
And one for offline renders only: Murmur processes audio in blocks of up to 65,536 samples, so an offline renderer asking for bigger blocks than that gets silence. Set the renderer's buffer size to anything at or below that and it bounces normally.
Three possible reasons, in order of likelihood. It is part of the full version, and section 2 lists which ones. It is inert because of another setting: DENSITY dims off the HZ clock, SIZE and SHAPE dim while REPEAT is on, POS dims in SLICE mode, SLICES dims while the boundaries are authored rather than detected, the grain filter's cutoff and resonance dim while its MODE reads Off, COHERE dims when the sample has no confident pitch (it still moves, and still pulls gently), and MORPH dims until a second layer is loaded. Or it only exists in some conditions: GLIDE at MONO, layer B's SLICES dial once the B link is off, and the AB / A / B buttons and PIN, which need a morph layer before there is anything to split. Hover it and the tooltip names the reason.
Some of the layer controls go flat rather than dim, which means the same thing: the choice cannot bite here. SIZE's button, the density ratio chip under DENSITY and WALK's B SEQ pill are all locked in the BLEND, RESONATOR and FOLLOW morph modes, because a grain in those three reads both layers through one window and belongs to neither. A routing row's button is locked for the same three modes, and also for a destination that is not per-layer whatever the mode. Nothing you set is lost while a control is locked; sections 5 and 8 have the detail.
If you are running the free version, the browser only lists presets that will sound the way they were made. A preset that leans on MORPH, COHERE, WALK, REPEAT, the FX rack, the two grain processors, ENV 2 or ENV 3, or any modulation route would come out thin and wrong rather than obviously locked, so it is left out of the list rather than offered and broken. The nine starter templates are all free-safe, so the free browser is never empty.
Nothing has happened to the file. It is still in your preset folder, it still holds everything it always did, and it comes back the moment a key is entered. Loading is not blocked either: a preset someone sends you opens by any other route with its state whole, and plays with its paid parts inert.
Expected, and nothing is lost. In the free version the second layer's violet lane is not drawn and the grains all come from the first sound, but the sample is still inside the session file. A free instance cannot eject it or overwrite it, precisely so that it survives. Enter a key and the lane reappears on the next repaint, with the MORPH dial exactly where you left it. The same is true of a WALK pattern: it stays in the file, and the step readout says nothing is walking rather than pretending.
Release installers are notarized and open cleanly. If you were handed a dev build, Control-click it in Finder and choose Open, then confirm. You only do that once. Alternatively use System Settings, Privacy and Security and click Open Anyway.
Samples are resampled when they load, not when the device rate changes. If you switch your interface between 44.1 and 48 kHz mid-session, reload the sample or reopen the project so it is prepared at the new rate.
That is how it behaves: the step clock is free of the transport's play state, so the phrase carries on while you are stopped. Let go of the note to silence it.
Three things drive it, in order. Drop VOICES: 16 voices of dense granular is a lot, and most parts need four. Drop DENSITY, since every grain is real work. Then look at the FX rack, where REVERB with SHIMMER up is the most expensive module. Switching a module off with its header dot takes it out of the chain completely.
Every preset saved by this version carries its audio inside the file. Presets saved by older versions from samples longer than 60 seconds referenced the sample on disk instead, and when that file is missing, Murmur names it over the display. Drop the file back on the display and re-save the preset to embed it for good.
Email support@soundslikeyoumusic.com. Tell us your operating system, your DAW, and the version number, which you can read by hovering the MURMUR wordmark in the header.
Found a bug, or want a feature? Tell us at soundslikeyoumusic.com/murmur/feedback. The changelog for every release lives at soundslikeyoumusic.com/murmur/changelog.
Murmur ships with third-party components: the VST3 SDK, the FLAC, Ogg Vorbis, zlib, libpng
and libjpeg libraries bundled with JUCE, and four font families under the SIL Open Font
License. Their full notices are in THIRD-PARTY-NOTICES.txt, shown by the macOS
installer and installed alongside the Windows standalone. VST is a trademark of Steinberg
Media Technologies GmbH.