Press a piano key gently and the hammer moves slowly, the string vibrates softly, and you get a quiet, warm tone with a gentle onset. Press the same key hard and the hammer strikes differently. More energy goes into the string, the attack is sharper, and the high frequencies are louder relative to the fundamental. Same pitch, different sound. The character changes in ways a volume knob can’t replicate.
A single recorded sample can’t capture that variation. You can play a recorded piano note at lower volume to simulate soft playing, but the timbre stays fixed. What changes between a soft and a hard strike on a real piano isn’t just loudness: it’s spectral balance, onset shape, and sustain character. Adjusting playback volume gets the loudness roughly right and leaves everything else wrong.
Velocity layers solve this by recording the same note at multiple playing strengths and assigning each recording to a range of MIDI velocity values. When someone plays a note softly, the sampler plays the soft recording. When they play hard, it plays the loud, bright one. The instrument responds to dynamics the way the real source does.
What velocity is and what it measures
In MIDI, velocity is a number from 1 to 127 that travels with every note-on message. On a piano or keyboard controller, it represents how quickly a key was depressed, which correlates with how hard the key was struck.
The number itself is just a value. What it means to the sampler depends on how the instrument is set up. A soft velocity might be anything below 45. A hard velocity might be anything above 90. The sampler watches the incoming value and plays whichever sample is assigned to the range it falls into.
Some controllers let you adjust the velocity curve so the response feels more or less sensitive. A linear curve maps controller force evenly to MIDI values 1-127. A convex curve clusters values toward the soft end so you have finer control at low dynamics. For the purpose of building a sampled instrument, the key point is that MIDI velocity is a discrete value the sampler uses to choose between recordings.
Recording multiple dynamics
The goal is a consistent set of recordings that represent different playing strengths without other variables changing between them.
Before the session, settle on your layer count. Three is enough for many instruments: a soft, a medium, and a hard. Five gives more gradation. More than seven layers has diminishing returns for most material and noticeably increases file sizes, so unless the instrument is a solo piano or close-mic’d acoustic instrument in an exposed context, three to five is the practical range.
Use the same acoustic environment and mic position for all layers. If you’re recording an acoustic instrument, don’t move the microphone between the soft session and the hard session. The dynamic change should come from playing intensity alone.
Name each recording to identify both its pitch and its dynamic level, for instance piano_C_soft.wav, piano_C_med.wav, piano_C_hard.wav. With hundreds of files across a full-range instrument, ambiguous names create slow, tedious problems later.
Keep levels consistent within each layer. If your soft layer is pianissimo at C3 and mezzo-piano at G4, the instrument plays unevenly across the keyboard. Pick a target dynamic for each layer and hold it throughout that layer’s session.
Record the full note, including the decay and release. At high velocity, many instruments have a longer, louder release than at low velocity. Cutting off the tail produces a click or an unnatural envelope shape. Let each note ring to silence, or close to it.
Leave a short silence (half a second is plenty) at the start of each file. It gives the pitch detector clean material and makes manual trimming easier if anything needs adjustment.
How samplers divide the velocity range
With three recorded layers, the sampler needs to know which recording plays at which velocity. A simple equal split looks like this:
Soft: velocity 1–42
Medium: velocity 43–84
Hard: velocity 85–127But equal thirds isn’t always the best choice. Most playing happens in a narrower dynamic range, so you might want more gradation in the middle:
Soft: velocity 1–30
Medium: velocity 31–75
Hard: velocity 76–127The split points are a mix of technical and musical judgment. Play the instrument after setting them and adjust until the transition between layers lands at the right musical moment. A soft-to-medium shift that fires too early makes the instrument feel tense and overresponsive at low dynamics. One that fires too late makes everything feel soft until suddenly bright and loud.
Some players add a velocity crossfade at each boundary, blending adjacent layers over a small velocity window rather than switching abruptly. Whether the blend sounds better than a hard switch depends on the material. Instruments with clear timbral changes between layers (brass, acoustic piano) often benefit from a clean switch that reflects the physical reality. Instruments where the dynamic transition is gradual (bowed strings, some organ stops) often sound more natural with a crossfade.
The SFZ format makes this structure explicit. Each region in an SFZ file can declare lovel and hivel parameters alongside lokey and hikey:
<region>
sample=samples/piano_C_soft.wav
lokey=60 hikey=64 pitch_keycenter=60
lovel=1 hivel=42That region plays only for notes 60 to 64 struck softly (velocity 1 to 42). A second region for the same keys covers medium velocity, and a third covers hard. The text-file format lets you read the entire mapping and change split points by editing numbers directly, without launching a separate editor.
Velocity layers vs round-robin
These two techniques get conflated because they both involve multiple recordings of the same note, but they address different problems.
Velocity layers address dynamics: the same note sounds different at soft and hard playing strengths.
Round-robin addresses repetition: the same note struck repeatedly at the same velocity sounds mechanical when the sampler plays the identical recording every time. The “machine-gun effect” is the most common description. Round-robin cycles through several recordings of the same note and plays a different one each time, so repeated hits don’t all land with precisely the same waveform.
A thorough instrument can have both. Several velocity layers, and within each layer, multiple round-robin alternatives. The combinations multiply quickly: five velocity layers with three round-robin samples per layer means fifteen recordings per note. That scale makes sense for a seriously detailed instrument; for most work, getting the velocity layers right is the priority.
How many layers you need
Acoustic piano and orchestral instruments tend to need more layers than electronic or synthesized ones. A piano has dramatic spectral changes across its dynamic range, and listeners are familiar enough with the sound to notice inconsistencies. Five layers is a practical floor for exposed piano parts. Three gets you through tracks where the piano sits lower in the mix. String sections need care at the soft end of the range, where pianissimo playing has specific bow-pressure character that medium-layer sampling misses.
For percussion, three to four layers on the most-used elements (kick, snare, hi-hat) covers the range from ghost strokes to full hits. One layer is fine for less prominent drums where velocity shading isn’t audible in context.
Synth patches with sustained, evolving textures usually don’t need more than one layer. Dynamic variation in those sounds comes from the synth engine rather than physical force, and a slow-attack pad doesn’t reward velocity sensitivity in the way a struck acoustic does.
The return from additional layers narrows as you add more. The difference between one and two layers is the most audible improvement in almost every case. The difference between five and six is often imperceptible unless you’re specifically comparing them.
Hardware module support
Velocity layers come up most often in software sampler contexts, because they require an instrument format that can store the layer mapping and a player that reads velocity from the incoming MIDI stream.
The Rample from Squarp Instruments is one of the few Eurorack modules with explicit velocity layer support. It assigns up to 12 sample layers per voice and can trigger them by velocity, round-robin, or random selection. In practice, getting velocity into the Rample requires a MIDI-to-CV interface that outputs velocity, since raw gate signals don’t carry that data.
Most simpler Eurorack trigger players fire at a consistent amplitude regardless of how the gate arrived. For those modules, velocity layers don’t affect the playback side, but they still make sense in the source instrument for use in software players or velocity-capable hardware.
Export formats and what they carry
If you build a velocity-layered instrument in SampleStack’s multisample mode, the software export formats all preserve the layer data.
SFZ stores each velocity range as lovel/hivel on individual regions. The mapping is human-readable and editable. The SFZ instruments guide covers the format, compatible players, and how the layer structure looks in practice.
SoundFont 2 carries velocity layers through its instrument and sample hierarchy. Decent Sampler’s XML format maps them to its group system, which the free Decent Sampler plugin reads directly. Ableton Sampler’s native format drops the zones into the device’s velocity range controls so they’re immediately visible and adjustable in Live.
The Disting NT supports polyphonic multisample playback through its Poly Multisample algorithm, but the playlist format doesn’t include velocity zone declaration. It plays one sample per pitch zone regardless of incoming velocity.
Before you export, SampleStack’s built-in format matrix shows what each format carries from your instrument. Velocity layers appear in the matrix for the software formats, so you can confirm what’s included before the export.
Tips for cleaner layers
Give the pitch detector clean, pitched material. Samples with a clear, stable fundamental map accurately. Noisy or inharmonic sounds need manual zone placement.
If you’re recording a hardware instrument into a DAW, use a MIDI controller to trigger the same note repeatedly at each velocity level and record each take to a separate file. This removes inconsistency from acoustic playing variation.
Keep the same monitoring level when auditioning the finished instrument. Layer transitions can sound abrupt at high monitoring volume and smooth at low volume. Pick a reference level that matches how the instrument will be used and check across the full velocity range there.
After mapping and before export, play the instrument chromatically from the bottom to the top of its range at each velocity level. You’re listening for uneven level between adjacent notes at each layer, which usually means a zone boundary is in the wrong place or one recording was louder than its neighbors.
If you built the instrument in SampleStack, SampleStack handles zone assignment and velocity range setup visually, which makes it easier to catch those level inconsistencies before they end up in the exported instrument.
Frequently asked questions
What’s the difference between velocity layers and round-robin?
Velocity layers are separate recordings of the same note at different playing strengths. Round-robin is separate recordings at the same strength, used to prevent the machine-gun effect when a note repeats rapidly. They solve different problems and can be combined.
How many velocity layers do I actually need?
Three is a workable starting point for most acoustic instruments. Five is common for piano. For synth pads or evolving textures where dynamics don’t change the character of the sound, one layer is fine. The jump from one to two layers is the most noticeable improvement; beyond five, the gains narrow quickly.
Do velocity layers affect file size?
Yes. Each layer is a separate audio file, so a three-layer instrument uses roughly three times the storage of a single-layer version. SFZ and SoundFont 2 keep all samples as individual WAV files, so the file count scales directly with layer count.
Which formats support velocity layers?
SFZ, SoundFont 2, Decent Sampler, and Ableton Sampler all carry velocity layer information. SFZ makes it the most explicit: each region in the text file declares lovel and hivel to define its velocity range. SampleStack’s export matrix shows what each format carries before you export.
Can I use velocity layers on hardware Eurorack samplers?
Some modules support it. The Rample assigns up to 12 layers per voice and triggers them by velocity, round-robin, or randomly. Most simpler Eurorack trigger players don’t receive velocity data and fire samples at a consistent amplitude regardless of gate strength, so velocity layers mainly apply to software formats and MIDI-driven hardware where velocity is present in the signal.