Record one continuous take of a single drum. Any WAV loads — 44.1 up to 96 kHz, 16, 24 or 32-bit float, mono or stereo. 48 kHz / 24-bit is a good default.

Mono is the usual choice for one drum. A stereo take dropped on its own stays stereo all the way through to the export. The one exception is a take that arrives as the close mic of several — a whole instrument, or a multitrack session: there the close mic is folded to mono, because it is the signal everything is measured and cut from and the width belongs to the overheads and the rooms, which keep every channel you recorded.

Play 7–9 hits at the same dynamic — these become that velocity layer's round robins — then step up and play the next 7–9. Work from very soft up to as hard as you play, so the whole range is covered.

Let every hit ring out. Never damp or mute a tail: the slicer keeps the natural decay and fades each sample just before the next hit for you.

That's it — drop the WAV below. Aiming for a specific count? See the table in the full guide.

What this does

Record one long take of a single drum — soft hits building to hard — and this turns it into a complete set of velocity-layered, round-robin samples: trimmed, aligned, tuned and named, ready to map in any sampler.

Nothing else automates this part. Auto-samplers drive an instrument over MIDI and record what comes back — which is no use at all when the instrument is an acoustic drum and the trigger is a human arm. Mapping tools want your samples already cut. This is the step in between: it takes the take you played and hands you the finished set.

It all runs in your browser. Your audio never leaves your machine.

The app checks itself as it starts — a handful of measurements with a known right answer. If one of them fails, or something throws an error, it says so and offers you a report: the version, your browser, and what broke. You read it first and you decide whether to send it, by copying it or opening a mail. Nothing is ever sent on its own, and the report contains no audio, no file names and nothing about the material you are working on.

Every section folds. Click any numbered heading to collapse that step down to its title — so you can close what you're done with and keep the step you're working in front of you. ☰ Collapse all at the top folds or opens them all at once. Each section remembers its state between visits. The deepest controls (global tune, the global start nudge and downtune quality, groove shaping, the demo's humanise & ring, the output format) sit in small ▸ sub-panels inside their section, one click away when you want them.

1 · Record the take

Tune first: the 🎯 Drum tuner (top of the page) is a realtime Tune-Bot-style tuner — start it, hit the drum, and it reads the ring's strongest tone as Hz / note / cents plus the overtone. Tap near each lug to even them out, or set a Target Hz to tune to an exact pitch. Nothing is recorded.

One continuous WAV, mono or stereo. Every format the app exports also loads: 44.1–96 kHz at 16, 24 or 32-bit float. 48 kHz / 24-bit covers it for most work — higher rates mainly earn their keep if you plan heavy pitch work. Play a run of hits at one dynamic (those become that layer's round robins), then step up to the next. Go from very soft to as hard as you play, and let everything ring out.

How many depends on the drum:

The snare earns the most work: it has the widest dynamic range (ghost notes up to rimshots) and it is the drum where repeated identical samples — machine-gunning — give the game away fastest. The kick needs the least: its useful dynamic range is narrow and round-robin variation is far less exposed.

2 · Load it

The load area takes four things:

• One continuous take (a WAV) — the classic path: it gets sliced by the analyser below.
• Several separated sample WAVs at once — select or drag a batch and each file becomes one slice, skipping detection. This is the way to re-open a set of already-cut samples and re-layer, round-robin, tune, add effects, or re-export them.
• A whole folder of sample WAVs — drag the folder onto the load area or use choose a folder; any sub-folders are searched too.
• A .zip of sample WAVs — same as above, straight from a zip. (A zip is auto-detected as either a sample pack or a saved session.)
• A saved .slicer.zip session — return exactly where you left off.

Only uncompressed PCM WAV decodes in the browser (44.1–96 kHz, 16 / 24 / 32-bit float, mono or stereo). MP3, AIFF and FLAC won't — export them as WAV first. If a file is refused, the load area now tells you exactly why. No recording handy? The demo buttons under the drop zone load a synthesised take — tom, snare, kick, ride or percussion — so you can try the whole workflow first.

3 · Settings

• Instrument name — becomes the file-name prefix (YourSample_01.wav). The app fills it in from your file or folder names; correct it here or at the export.
• Take type — AUTO measures how long your hits ring and picks SHORT (toms, snares) or LONG (cymbals, effects, reverb), then shows you which it chose. Override only if it guessed wrong.
• Advanced — detect threshold, minimum gap, max sample length, stray rejection. The defaults are tuned against real takes; leave them alone unless something looks off.

4 · Analyse

Finds every hit, aligns each one precisely to its attack, and colours them green (softest) to red (loudest). Each sample automatically runs to just before the next hit and fades softly there — so nothing clicks, and no decay gets chopped.

5 · Review & pick

• A multitrack take — a WAV with three channels or more is not a stereo take, it is a session, so the import stops and asks which channel is which microphone. A WAV carries no names for its channels; that is the one thing it cannot tell the app, and guessing wrong would misalign a whole kit. The usual layout is filled in from the channel count (close mic, then overheads, close room, far room, and a third room if there is one) and any row can be corrected, set to mono or stereo, or switched off. Name the close mic — that name is the zone, because the close mic is the thing you hit, and it names the exported files.
• Or drop the takes as separate files — one WAV per microphone, which is how a DAW hands a session over. Drop them together and the same question is asked in a table: which file is which. Pick the instrument at the same time (tom, snare, kick, percussion, cymbal) and name each position; those names go on the exported files.
• Two close mics — a snare top and a snare bottom — are handled. The top is the master: it is the take that gets analysed, detected, sliced and shaped. The bottom is a position like the rooms, cut at the master's own points and never re-timed, and it has an invert Ø tick. That is polarity, not phase: a mic under the drum looking up at the wires produces a waveform that is upside down against the top mic, and summing the two partly cancels them. Multiplying by −1 is exact; a phase shift would be frequency-dependent and would smear the attack. Measured on a real pair: summed peak 0.27 without the tick against 0.29 for the top mic alone — the bottom was making it quieter — and 0.39 with it. Every position has the tick, because a room can need flipping too.
• Every other position then replaces the model for that mic, and the Source switch on each one flips back to the emulation so you can hear them side by side. They are cut at the close mic's own points, so every trim, every repair and every , . keypress moves them by exactly the same amount — and the time between them is never touched. The import measures how late each position arrives and tells you (an overhead 1.3 m up arrives about 3.8 ms after the close mic — that is the distance, and it is what makes a kit sound like a room), but it does not correct it: the relationship you recorded is the one you keep. Hundreds of milliseconds would mean the tracks are not from one pass.
• Whole take — include the full-length tracks (in the export card, off by default) — not a sample pack but the source, in the same zip as the samples. A Take folder holding the close mic under the name you gave it and every mic position you have switched on, each at full length, exactly as they were recorded against each other. Nothing is re-timed and nothing is trimmed to fit — channels are padded to the longest, because shortening one to match another is precisely the re-timing this exists to avoid. Drop the folder on a DAW timeline and every track lines up at zero.
• The same take also goes in interleaved, as one multitrack WAV, and that is the file to drop back onto the load area to open the whole session again with every microphone in place. Modelled positions are rendered over the whole take rather than hit by hit, so a recording that only ever had a close mic comes out as a real multitrack session — which is how you make multitrack material to work with. Verified by doing the round trip: 14 hits and three positions out, 14 hits and the same three arrival times back.
• Only the close mic is sculpted. Drum size, match EQ, stack, bend and the shaping tools exist to reshape a close mic, and imposing them on a real overhead would invent something the microphone never heard. Tuning and per-hit level DO follow, or a drum tuned down would carry an untuned room and the two would drift apart into different drums.
• Click a tile to hear it. A keeps it, D discards it. Show opens on Kept, so a discarded sample leaves the grid completely — it is not shown, not stepped onto by the arrow keys, not caught by a range selection and not audible. Nothing is lost: Discarded is a real view where you can listen to what you threw out, and A there brings it straight back.
• Pick several at once — shift-click for a range, ⌘/Ctrl-click to add or remove one, ⌘/Ctrl+A for all, Esc to clear. A or D then acts on the whole batch as a single undo step. A batch only ever covers what the current Show is displaying, so nothing off screen can be changed by a key you pressed while looking at something else.
• Reorder — drag a tile by its ⠿ grip, or nudge it with ◀ ▶. The order sets the export numbering.
• Tune — the +/− on each tile, or the + and − keys, in cents. Hold shift for steps of 10. With several selected they all shift by the same amount and each keeps its own offset.
• Auto-tune (in Tuning) — a drum has no single pitch: its head rings at inharmonic modes, which is why ordinary pitch trackers jump around on toms. This works the way hardware drum tuners do instead: it finds each hit's fundamental drum tone (the strongest low spectral peak of the ring, after the transient), takes the take's consensus, then re-measures every hit only near that tone and tunes the small offset away — so corrections stay small by design and an octave jump can't happen. Tune to either MATCH (the take's own tone — keeps the drum's natural note, just consistent) or NEAREST NOTE (snaps that to the closest semitone). Hits with no clear tone are left alone, so an inharmonic take (snare, hats, most cymbals) reports nothing to tune rather than being mangled. Undo reverts it. It works well on some drums and not on others — always finish by ear: listen, then fine-tune by hand where needed.
• Size L or XL also shows each hit's detected pitch, its note, and how many cents it sits from the take's reference — so you can see what's worth tuning by hand. Hits flagged beyond ±250¢ read unreliably (drums are inharmonic); don't trust those numbers.
• Tighten starts — the Starts row reports how many samples pass before the hit actually arrives, worst and typical. A take sliced here lands on 1–3. An imported instrument, or a session cut by a version before v1.210, will not: it carries dead air on some samples and a chopped attack on others. The button re-runs the current start analysis over every sample and fixes both at once. It is offered in the Quick setup row too, on the same terms as Sort by velocity, whenever the material needs it. Stunt doubles and any start you moved by hand are left alone.
• If a start comes out wrong, the app repairs it — and if it cannot, it says so. The analysis aims for 1–3 samples of head; on the six reference takes it holds that on 476 hits out of 477. Past 5 it is not a near miss, so the app measures the head in front of that sample and removes it — but only after checking the move: what it would discard has to sit under −25 dB of the hit’s own peak, and the head really has to come down. Anything else is put straight back, because cutting into an attack is a far worse mistake than leaving a few samples of silence. Measured by breaking every cut on all six takes and asking it to undo the damage: it repairs 86–89 %, lands a median of 2 samples from where the analysis had it, leaves the first exported sample about 8 dB quieter than the analysis did, and has never cut late.
• Whatever it could not repair safely gets a red bar naming how many and how bad, a ⚠ flag, the START zoom, and the worst one selected and scrolled to — , and . then move its start one sample at a time, so in the worst case you can finish it by hand. Either way it goes into the self-check report so it can be fixed properly; sending that report is still your decision, made after reading it. And none of this is ever said about material the analysis did not just cut — it runs on a fresh Analyse or on Tighten starts only, because an imported instrument starts wherever its author cut it, and calling that a bug in this app would be a lie.
• Hold B — hear the selected sample raw: as recorded, with no tuning, no effects, no mics and no level sculpt. The other half of "is this better?".
• Home and End jump to the previous / next velocity layer, so you can get from the softest end of a set to the hardest without holding an arrow key through seventy samples.
• Size and Zoom sit together in the view strip above the tiles, with Show, the state of the set, and the Starts readout. Everything that decides what you are looking at is in one place. Waveform only hides the text rows and leaves the pictures, for going through a set quickly by ear.
• Zoom DETAIL and START reveal the onset nudge. The single arrows move the start one sample (or press , and . — hold shift for ten), the double arrows move it 0.2 ms, and clicking the sample count puts it back where the analysis placed it. Every waveform is drawn over a baseline so you can see whether a start sits at zero, and the close-up views print lead — how many samples pass before the hit reaches −40 dB.
• Play it — number keys 1–0 trigger five dynamic zones with random round robin.
• Order — PLAYED keeps the recording order, BY VELOCITY sorts soft → loud, and ANTI M-GUN goes one further: soft → loud, but the round robins inside each layer are re-sequenced by spectral difference so consecutive repeats sound as unlike each other as possible — the machine-gun effect dies. It also warns if a layer contains two nearly identical hits. Drag tiles for a fully manual order.
• Play all kept runs through every kept sample in order; press it again to stop. On a long take that is mostly decay, so set Length to 1/2, 1/5 or 1/10 to get through it quickly — it shortens the audition only, never the exported files.
• Grooves — six patterns, each leaning on a different part of the set: 1 backbeat, 2 ghost-note 16ths (the soft round robins), 3 a ppp→fff build with a fill, 4 shuffle, 5 half-time with room to ring, 6 linear funk. Plus CRESC, ROLLS, a swelling BUZZ roll and RANDOM. All humanised, and all using the exact tuning and fades that will be exported.

6 · Map & name

• Plain — name_01 … name_NN, soft to loud. A straight series most samplers will map across the velocity range for you.
• Velocity × round robin — tick it to name files name_VV_RR (velocity first, then round robin): the layout Kontakt, SFZ, Decent Sampler and the Pearl Mimic Pro all expect. Set round robins per layer to however many hits you played per dynamic. The Layer Map then warns about layers with too few round robins, or a big loudness jump between layers.
• Also write an .sfz — hands you a finished instrument rather than a pile of files. The velocity layers and round robins are already mapped, so sfizz, Decent Sampler, Bitwig, Cakewalk or any other SFZ host opens it and plays. Set MIDI note to wherever the drum belongs on your kit. Ticking this always produces a zip, since the .sfz and its sample folder have to travel together.
• & a DecentSampler .dspreset — the same instrument as a native DecentSampler file (free player). Tick it alongside or instead of the .sfz; both share the same MIDI note and both ride in the zip next to the sample folder. DecentSampler can also import the .sfz, but its own .dspreset loads cleaner.
• 📄 Spec sheet — a printable one-page document of the finished library, saved as an HTML file you can open anywhere and print (or save as PDF) with Ctrl/Cmd + P. It lists what the kit contains, the velocity map (layer, MIDI velocity range, round robins, level range, file names), the kit's tone averages, and every sample with a waveform thumbnail, length, peak, RMS, tuning and any QC note. Everything is measured from the finished slices — effects and tuning included — so it documents the files you are shipping, not the raw take. Send it with a library, or read it in six months when you have forgotten what you did.

7 · Analyse & build

Its own step below Map & name — a read on the whole set, all from the same loudness (rms dB) the colours and layers use, plus the tools to reshape it and to build the samples you didn't record:

• QC pass — scans every kept slice and flags the suspect ones so you audition only those, not all of them: clipping, a possible double-trigger (two hits caught in one slice), a very short slice crowded by the next hit, a level outlier sitting far from the rest of its velocity layer, and a late start — more than 5 samples of head before the attack. Flagged tiles get a ⚠ corner badge (hover for the reason); click any item in the panel to jump straight to that sample and listen. It never deletes anything — a flam you meant to play and an accidental double-trigger look alike in the data, so QC points your ear and you decide.
• Dynamic map (default) — every kept hit plotted in recording order against its loudness. A clean soft-to-loud take climbs steadily; plateaus or big jumps show where coverage is thin or uneven. It's also an editor: drag any dot up or down to set that hit's level (a faint stem shows how far it moved from where you played it). Drag across empty space to box-select a run, then move them together. The buttons below the map: Even out layer snaps a velocity layer's hits to their shared median so round robins sit together; Draw sweep lets you sketch a target level and the hits you drew over conform to it — only that span, so a short flat line levels one jumpy patch and leaves the rest of the take exactly as it was (drag right across the plot to reshape the whole thing, as before); Reset levels returns the selection (or everything) to as played. It's all non-destructive — the level flows into the colours, the layers, QC, the demo and the exported WAV, and Undo takes it back. A hit whose level you changed shows a small ▲/▼ amount on its tile.
• Velocity histogram — how the hits spread across loudness, with the count on each bar and dashed velocity-layer edges. Clusters and gaps show at a glance.
• Layer balance — round robins per layer with each layer's dB range; ⚠ marks a thin layer (may machine-gun) or a big loudness jump between layers.
• Decay times — how long each sample rings (peak → −14 dB). A tight cluster means consistent tails; a low outlier may be choked or cut short. Punch — peak against rms for every hit: sitting high above the pack is a punchy transient, and any dot on the red 0 dBFS line is clipping.
• Spectrogram — the selected sample as time × frequency, so you can see what a waveform hides: flat horizontal lines are ring modes (the boxy tone you may want to tame), a haze under the tail is noise or bleed from another drum, and two round robins tuned apart show their lines at different heights. Click any tile to switch sample.
• RR overlay — every round robin of one velocity layer laid on top of each other. A sampler fires these in turn for the same key at the same velocity, so they ought to be near-identical — this view makes the one that isn't jump out. The top panel overlays the waveforms aligned at their start, at true relative level with polarity kept: a round robin that sits louder or quieter, one whose attack tick lands late (a loose trigger), or one flipped upside-down (a polarity slip) is obvious. The bottom panel overlays their spectra on a log-frequency axis, where a round robin tuned apart shows its peak at a different height. It reads the layer of whichever tile is selected — click another tile to move to its layer — and flags a likely polarity flip or an off-tuned round robin underneath.
• Tone fingerprint — five perceptual descriptors of every hit on their own lanes: Brightness (spectral centre), Warmth (low-mid weight), Punch (crest / transient), Sustain (ring length) and Grit (tonal ↔ noisy). See your kit's character and its spread at a glance, and spot a hit that sits far from the pack — click a tile to spotlight it on the lanes. Relative within the kit, not an absolute measurement.
• Scroll to zoom the dB axis on the first two views, double-click to reset. The stat chips read out detected / kept / discarded counts, dynamic span, quietest and loudest kept, median level, noise floor, take length, and the source rate / bit depth / channels.
• Tame clipping on export — an optional switch that pulls any slice reaching 0 dBFS back to −0.3 dBFS, so a hot take doesn't ship clipped samples. Off by default; the QC pass shows you which ones clip.
• ✦ Stunt Double Velocities — on the bar under the sample grid, because a thin layer is something you notice looking at the tiles. A hand-played take is always short somewhere: a velocity layer with only two or three round robins (which machine-guns on repeats), or a gap in the dynamic range you never hit. This builds the strokes you never played out of the ones you did, and every part of every new hit is real audio from this take — what is invented is the combination. Each new hit gets its own seed, and everything is drawn from it: the whole hit is detuned by a couple of cents by stretching (not a pitch shifter, which would smear the attack), the three bands shift a little in balance and decay — what a stroke nearer the centre, or harder into the head, really changes — the head comes from one stroke and the tail from another, slid into step with it and level-matched across the join, and the brightness follows the curve this drum actually shows when it is hit harder, fitted from your own hits. The bands are never detuned against each other: two bands overlap, so they carry the same frequencies, and sliding those two copies apart is how a flanger is built — which is exactly what the first build sounded like. One seed per hit means every microphone gets identical numbers, so the kit stays together. Nothing is added until you say so: Roll a new set, Hear one it made against the played one next to it, or play 🎮 Spot the fake — four hits at one level, one of them made up — and then Keep them. Kept hits are normal kept samples: they colour, QC, play and export like the rest, wear a ✦ on their tile and a ring on the map, and Remove them or Undo takes them back.

8 · Export

• Output format — collapsed to one line showing what will be written; open it to change the rate and bit depth. KEEP ORIGINAL converts nothing, which is what you want unless your target demands otherwise; recorded at 96k or 192k, it stays there. Most libraries ship 44.1 or 48 kHz at 24-bit. 16-bit is dithered. 32-bit float is the app's own internal format written straight out, unquantised — check your sampler takes it. 44.1 kHz + 24-bit is what the Pearl Mimic Pro Instrument Editor accepts — it only imports 44100 Hz files. The catch: the Mimic's playback engine runs at 48 kHz, so a plain 44.1 kHz sample comes out about 1.5 semitones sharp on the module. The open question is where that sharpening happens: the desktop Editor's preview may resample wrong while the actual module brain plays the 44.1 kHz sample just fine. So it is off by default. If a test on your real module shows samples playing sharp, tick Compensate pitch for the Mimic's 48 kHz engine: the app pre-warps the audio so it lands in tune on the 48 kHz engine while the file stays a 44100 Hz WAV the Editor imports. If the module already plays plain 44.1 kHz in tune, leave it off. Soft samplers (Kontakt / SFZ / Decent Sampler) read the rate from the file, so leave it off for those too.
• Whatever you pick, the pitch is exactly what you tuned here. A rate conversion rescales the sample count in step with the rate, so each slice keeps its duration in seconds — and pitch is duration. Verified across every rate and depth to within a hundredth of a cent.
• Five files or fewer download individually; more arrive as a zip.

Effects (optional — baked into the export)

An optional chain you hear live everywhere you audition — tile clicks, Play all, grooves — and that is printed into the exported WAVs. Use it only for character you do not plan to add later in your sampler. Everything is off by default; ▶ Test plays the selected sample through the current chain, All off clears it, Panic stops all sound and any ringing tail. The panel is split into Colour & space (on the shared master bus) and Per-sample shaping (baked into each hit, and what FX amount by velocity scales), each showing how many are on. Each effect's settings are collapsed — the checkbox beside the name switches it on/off; click the name (or the ▸ arrow) to open its controls when you want to dial it in.

• Presets — the bar at the top loads a whole chain at once. Pick a factory starting point (808 Drop, Lo-Fi Tape, Gated Room, Vintage Drive, Robot Ring, Wah Funk, Spring Surf, Cathedral) then tweak. 💾 Save… stores the entire current chain — every effect and setting — as your own preset (kept in this browser); Delete removes one of yours.
• 🎤 Record the take yourself (beta) — in Load take. Record straight into the slicer from an audio interface, with every mic at once: the close mic that gets sliced, plus real overheads, close room and far room, each mono or stereo. Press Detect channels to see how many inputs the browser will actually give you (more than two needs a real multi-channel interface), assign each mic to its channels, and record. Because the tracks are captured together they are sample-aligned by definition, so the hits found in the close mic cut every other track at exactly the same points — there is no drift to line up. A recorded position replaces the emulation for that position and nothing else, so recording only the overheads and letting the model build the rooms works fine. On a recorded mic only the per-hit gain and the tuning are applied, never the close-mic sculpting tools (shape, size, match EQ, stack, bend) — those exist to reshape a close mic, and imposing them on a real overhead would invent something the microphone never heard.
• 🎙 Overheads & room mics (beta) — in Export. Each position holds either your own recording of it (if you captured one in step 1) or a model of it built from the close sample, and a Source switch on that mic flips between the two so you can hear them side by side. A recorded position wins by default, and nothing is discarded either way. What follows describes the model. Each pair has a Pair and a Pattern. Pair is real geometry, not a preset name: AB spaces the mics and images purely by arrival time (biggest and widest, least mono-safe, and the only one where the Spread slider is yours); XY puts them in the same place crossed at 90°, so there is no time difference at all and it folds to mono perfectly, imaging by level alone (±4 dB at full pan); ORTF is 17 cm at 110° and NOS 30 cm at 90°, both giving time and level together. Pattern changes how much room the mic hears as well as its tone — an omni gathers three times the diffuse energy a cardioid does, so the identical position comes out about 5 dB wetter; figure-of-eight rejects the sides and picks up the back in reverse polarity. Your samples are close mics; a real kit is close mics plus overheads plus room mics, and that blend is most of why a recorded kit sounds like a room instead of a sampler. This builds the extra mics from the close sample acoustically: a shoebox room (TIGHT / STUDIO / LIVE / HALL, with Size and Damping on top), mirrored copies of the drum for every reflection path, and cardioid pairs aimed at the kit. Because each mic has a real position it arrives late by exactly the time sound takes to travel there — 3–4 ms for the overheads, ~7 ms for the close room, 14 ms+ for the far room — and the two channels of a pair genuinely differ because they stand in different places. Position in kit is a real place, not a pan pot: move a tom off centre and it reaches one overhead first, exactly as it would. Spread is the mic spacing, so widening it widens the image honestly. Low cut matters more than it looks — distance mics gather low end a close mic never had — and Squash is the room compressor whose pumping tail-lift is the classic room sound. Mix it all with the faders and press space to hear the blend. Then pick how you want them, at the top of the panel: as separate sample sets writes each mic to its own folder — called what you called that mic, with the close samples in a folder named after the close mic itself — numbered to match the close files one-for-one so they layer onto the same pad, leaving the close samples dry for you to balance later; baked into the samples mixes the blend straight in, giving one finished stereo sample per hit — and that one also applies to .mci and .dsnd, and to the showcase demo (whose own Room is bypassed, since the mics bring a real one). Tick both if you want both. Tick neither and none of it is used anywhere — those two boxes are the on switch. Marked beta: the model is measured and validated, but a model is what it is — which is exactly why the Source switch exists once you have recorded the real thing.
• The effects — reverb (room/plate/hall/gate/spring), delay (echo / ping-pong / tape, tail up to 10 s), chorus, flanger, phaser, drive, bitcrush, ring mod, auto-wah, tremolo, octave fuzz, resonator and a formant/vowel filter. Each has its own mix so you can dial it in.
• MINE reverb — your own room (or any sound) as the space — pick the MINE preset and either 🎤 Record room (allow the mic, then give one sharp hand clap — the app trims the capture to a clean impulse) or 📁 Load IR to use any WAV as the impulse: a cymbal ring, a slammed door, a synth chord. Your drum then reverberates through that sound. Length crops the tail, Tone darkens it, and the impulse is saved inside the session so it comes back exactly.
• Shape — splits every hit into three components and lets you rebalance them: Snap (the stick attack), Body (the tonal ring of shell and head) and Air (wires, sizzle, room air), each −24…+12 dB. More punch without EQ, a fatter or tighter drum, a darker or wirier tail — and at 0/0/0 the sound is untouched. Baked per sample, like Pitch bend.
• Ring tamer — kills the one overtone that rings too long: the boing over a tom, the boxy 400 Hz on a snare, the ring you would reach for tape to fix. Press 🔍 Detect and it finds the mode that still stands out of its neighbours late in the decay — deliberately searching above the drum's own tone, because that is the note, not the problem — then set Depth by ear. What makes it different from notching the same frequency in the Graphic EQ: that cut is on all the time, and it takes the drum's body with it (which is why taped drums sound dead). This notch is dynamic — it stays open while the stick and the broadband body are sounding, and only closes once the ring is carrying the sound alone. Width is the notch width in octaves, Keep attack holds it fully open for the first few milliseconds. It ignores FX amount by velocity on purpose: a soft hit's ring is just as ugly as a loud one's.
• Drum size & material — a small physical drum modeler. Set a target Diameter (6–50″) and Depth (1–50″) in inches: a drum's tone is roughly inversely proportional to its diameter, so doubling the target drops the ringing body an octave (and lengthens it) while the stick attack stays exactly as recorded — a 12″ tom can become a 20″ without the slowed-down-tape smear. The sliders start at the app's estimate of the current drum (shown under them), so moving them resizes from there. Shell material — WOOD (neutral), ACRYLIC (loud/punchy), FIBERGLASS (tight/dry), ALUMINUM & STEEL (bright, metallic, longer ring), BRASS (warm + bright) — applies that shell's tonal signature and, for metals, a little extra ring. Dampening is moongel/tape: it shortens the ring for a tighter, drier drum. Ring boost is the same tail-sustain lift as the standalone effect, here per drum. Snare mat (beta) adds synthesized snare-wire buzz gated by the drum's own envelope, so any tom — or even a kick — picks up a snare-ish rattle. All of it bakes into the exported samples, and the attack never smears.
• Match EQ — make your drum's tone sit like a reference. Load any sample whose sound you're after (a commercial one-shot, an old favourite); the app compares its spectrum to your kept hits and computes a correction EQ (capped at ±12 dB, loudness untouched), then bakes it into every sample. Amount scales it from subtle to exaggerated (200%). The learned curve is saved with sessions and effect presets — so a matched tone can be stamped onto any future kit.
• Stack layer — the studio "reinforcement layer": load a second sound (a sub-boom, a clap, a noise crack, a bigger drum) and it's peak-aligned to every hit's transient and mixed in underneath. Level sets how loud it sits, Tune pitches it to the drum (e.g. a sub to the kick's note), Tone tilts it dark (a sub) or bright (a snap), and Align peaks keeps the stack phase-tight. Follow vel makes the layer track each hit's own dynamics so the balance holds at every velocity (at 0% the layer sits at a fixed level and flattens the kit), and Vary nudges every hit's layer a little in pitch and level so repeated stacked hits never sound like clones. Build a kick+sub or snare+clap without leaving the app; the layer is saved in the session. Baked per sample.
• Graphic EQ — a 31-band ⅓-octave graphic EQ (the same ISO centres as hardware units), ±12 dB per band. Drag a bar, or sweep across several to paint a curve; double-click zeroes a band, Flat resets all. Applied as a zero-phase filter, so the response is exactly the curve you drew — no band interaction, no phase smear, and the white line shows the true applied response. Baked per sample.
• Pitch bend — a per-sample pitch envelope (the tom / 808 drop or a riser), baked into each hit. Bend sets the depth (far more range down than up), Start where the drop begins near the transient, Fall how fast it slides, Start pitch offsets the whole sample. Ring boost lifts the quiet low tail so the drop is clearly heard, and Ring length time-stretches that boosted low ring so it sustains longer — a long, natural boom. A stretched sample auto-trims any dead air at the end and fades out over up to ~3 s.
• Ring boost — the same tail lift as Pitch bend's Ring boost, but standalone — no bend needed. It upward-compresses the sample: the quiet decaying ring is pushed up toward the peak while the attack is only gently tamed (never ducked), so a dead tom or a choked cymbal rings on. Amount sets how hard — measured, a 70% setting lifts the tail about +17 dB while the attack barely moves. Baked per sample, and it follows FX amount by velocity.
• FX amount by velocity — at the bottom of the panel. A real drum doesn't just get louder when you hit it harder; it brightens and bends more too. Tick this and the amount of Shape, Match EQ, Graphic EQ and Pitch bend follows how hard each hit was struck — the loudest hit gets the full setting, ghost notes stay clean. Depth sets how far the amount falls away on soft hits (100% leaves the softest completely dry). The live-chain effects (drive, crush, reverb, delay…) stay even because they run on one shared bus — that is what keeps what you hear and what you export identical. Stack has its own Follow vel.
• Stacking is safe — effects run in series, so many at once would normally get too hot and crackle; the app auto gain-stages the effected signal so it stays clean. The trade-off is that a big stack plays a little quieter — raise the master volume to taste.

Save your work

💾 Save session writes a .slicer.zip holding your original take plus every decision — keeps, discards, tuning, order, nudges, and the whole effects chain. Drop it back on the load area any time to resume exactly, with no re-analysis.

↻ And you no longer have to remember to. Every edit is kept in this browser as you work, and so is the take itself. Come back after a crash, a closed window or a restart and the load area offers ↻ Pick up where I left off — one click, no finding the file, and you are exactly where you were: keeps, discards, tuning, nudges, order and effects.

Why keep a copy of a file you already have? Because a web page cannot reach back into your disk. It only ever receives the bytes you hand it, and it is never told where they came from — that is a security boundary of the browser, not a shortcut this app skipped. Chrome can be granted a lasting handle to a file; Safari cannot, and Safari is where this problem showed up. So the take is stored where it does no harm: on disk, not in memory, one take at a time, replaced by the next one you open, refused above 200 MB, and erased by Forget it.

If the take is gone but the decisions are not — a different browser, or storage cleared — open the same audio again and they still come back by themselves, with Start fresh instead right there if you wanted a clean slice. It has to be the same audio: sample rate, length, channel count and a checksum of the take all have to match, so another drum’s decisions can never land on this one. Nothing is uploaded, ever.

Keyboard shortcuts

If something looks wrong

• “This webpage was reloaded because it was using significant memory” (Safari) — a take is a big object: one minute of 44.1 kHz mono is about 10 MB once it is decoded, and a stereo take is held three times over (left, right, and the mono sum every measurement uses). v1.222 halved what is held — the decoded copy of the take was being kept alongside the working copy for the whole session, and read by nothing — so this is rarer now. It cannot be made impossible: no browser publishes the line it kills a tab at. What it costs you is nothing any more: your decisions and the take are both kept as you work, and one click on the load area puts you back exactly where you were. Long takes are still the thing to watch — the self-check panel reports exactly how much is being held.
• Sound gone silent — press 🔈 Reset audio (top of the page). It rebuilds the audio engine and confirms itself with a short beep. Safari in particular can strand a page's audio after the output device changes (headphones connect or drop, the display sleeps) or after long idle; the reset recovers every case a page can recover. If even the beep stays silent, Safari's own audio process is stuck — reload the page first, and only quit and reopen the browser if a reload doesn't bring it back.
• Hits missed — lower the detect threshold, or shorten the minimum gap.
• One long ring chopped into several slices — force take type LONG.
• Several hits stuck inside one sample (a long-ringing drum — rototoms, high-tuned toms) — fixed in v1.142: the detector now spots a stick attack by its high-frequency transient, so a strike landing on a still-ringing drum starts its own slice. If a few still merge, shorten the minimum gap; if hits are very soft, lower the detect threshold too.
• Quiet hits marked AUTO-STRAY — they were far quieter than their neighbours. Press A to keep one if it was real, or set stray rejection to 0 to switch the check off.
• Samples sound cut short — raise max sample length, or leave more space between hits when recording.

A realtime tuner (Tune-Bot style): start it, hit the drum, and the strongest tone of the ring is read out as Hz, note and cents — with the overtone beside it. Tap near each lug to even them out, or set a Target to tune to an exact pitch. Nothing is recorded.