Pocket: drum research and implementation brief

Research date: 2026-10-29. This is a synthesis of production teaching and sampler documentation, not an original transcription study or a claim to have auditioned the linked recordings. The app's original synthesized examples let the user hear the implementation choices directly. Lo-fi hip-hop is a broad practice, not one timing formula; Dilla's wider work is not interchangeable with contemporary study beats.

Sound and construction

Native Instruments' lo-fi production walkthrough starts with a restrained hip-hop skeleton, acoustic/vintage sources, regular hats, selective note displacement and softer secondary hits. Bandwidth reduction, resampling and a small amount of room sound are separate texture decisions. Its backbeat on beats 2 and 4 is normal backbeat time at the stated pulse; its loose use of “half-time” should not become a rule that every lo-fi pattern puts the snare on beat 3. A soft kick can still have a definite attack; dark hats need enough upper-frequency information to articulate time. Shorter tails leave space, while room and cymbal spill can bind break slices together. Excessive filtering can remove the snare's identity. Source and embedded before/after examples.

Pocket is relational

Ableton's groove manual separates timing, quantization, velocity and randomness, discusses extracting timing/volume from recordings and applying different grooves to individual voices. Thus a stable kick, displaced snare and differently swung hats are independent decisions. Random offsets can separate sounds that originally coincided; they are not a substitute for intentional placement. Pocket implements repeatable offsets and a straight-timing comparison. “Straight” keeps the same hits and dynamics, so the timing difference is isolated.

Humanize Your Drums collects producer demonstrations of alternative subdivisions, sound variation and fills. Triplets, quintuplets and septuplets suggest different feels; none establishes a universal Dilla percentage. The implementation uses explicit eighth- or sixteenth-note swing, per-lane offsets and per-note offsets. It does not claim to reproduce a particular artist.

Swing math used here: 50% is equal spacing; 66.67% places the second note of a pair at two thirds of that pair. At 80 BPM a sixteenth is 187.5 ms; 58% sixteenth swing delays every second sixteenth by 30 ms. Eighth swing shifts the offbeat eighth, with intervening sixteenths interpolated. Confusing these two bases changes the groove. A +12 ms snare offset is a design starting point, not a measured genre standard. Early snares can also create a compelling pocket.

Dynamics and phrase design

Native Instruments' drum programming lesson demonstrates soft ghost notes, contrasting accents, layering and end-of-phrase variation. More notes need not mean more force: a soft snare pickup should remain subordinate to the backbeat. Alternating hat levels creates an audible pulse even before swing is applied. Sparse kick syncopations can answer a strong downbeat. Fills work as punctuation rather than continuous rolls. Our two-bar suggestions keep backbeats recognizable and vary the second bar. Density changes supporting hits; it does not erase the skeleton. The generated numbers and patterns are our musical design choices, not copied MIDI transcriptions.

Sampling, breaks and cleanup

Ableton's slicing documentation distinguishes slicing by time, transients and markers from interpreting the musical content. Slices can be triggered in original order or rearranged, and the original timing can be retained. A detector finds likely boundaries, not isolated instruments: a slice may contain kick, hat, room and bleed together. Therefore Pocket offers sensitivity, manual boundaries, auditioning, role assignment and a separate original-order break sequence. It does not silently label every transient a kick or claim source separation.

Practical engineering choices: retain original decoded audio; trim non-destructively; show start/end in seconds; use short fades to prevent discontinuities; cap normalization gain; allow pitch, reverse and low-pass filtering. Silence detection is amplitude-based, so audition soft attacks before accepting it. A transient detector can miss soft ghost notes or split ringing snares; manual cuts remain necessary. Pitch changes playback duration (resampling), not independent time stretching. The original-order reconstruction places slices at their source-relative positions, scaled to a user-declared one/two/four-bar region. An incorrect bar count gives an incorrect BPM. Mixed slices cannot be unbaked.

Suggestions and listening experiments

Three original recipe families: Dusty backbeat (stable kick, modest sixteenth swing, slightly late snare), Loose pocket (straighter kick, early snare, eighth-swung hats), Sparse halftime (snare on beat 3, fewer supporting notes). These are presets to audition, not musicological categories. Only loaded, unlocked lanes are changed. Role assignments are user controlled. Timing offsets repeat identically each cycle. Candidate preview, Keep, Discard and undo protect deliberate edits.

Audition procedure: start the demo; compare Pocket/Straight without changing density; solo the hats to hear accents; mute hats to hear kick/snare relationships; compare 50/58/66% at both swing bases; lower ghost velocity; import a break and compare original-order slices before rearranging. Check that pitch-down lengthens slices and can overlap; shorten tails or enable lane choke if necessary. Test the rendered loop against live playback and across the wrap.

Classic pattern library

Added 2026-10-29: sixteen original two-bar templates in four groups: boom bap, old school, breakbeat and laid-back. These are style exercises, not transcriptions of named songs. Kick placement, ghost notes, hat density and second-bar turnarounds are authored explicitly rather than randomized. The Attack Magazine boom-bap lesson demonstrates the practical importance of velocity, separately treated hats and kick/snare timing. It informs the approach, not the exact note data here. Suggested tempos and offsets are starting points to audition, not historical measurements. Templates use the sounds assigned to each role, preserve locked and unassigned lanes and the break lane, and keep the current tempo unless requested. One-bar loops use the first bar; four-bar loops repeat the two-bar phrase. Straight comparison remains available. Preview, Keep, Discard and Undo apply to templates just as they do to generated suggestions.

Scope and limitations

Browser-local samples and portable project files; no copyrighted demo breaks. Initial release has six lanes, up to four bars, WAV export and project backup. No AI audio model, stem separation, tape emulation, independent time stretching, song arrangement, plugin hosting or MIDI hardware recording. Suggestions are explainable rules grounded in the sources above. Desktop Chromium is the automated verification target; other browser/device audio behavior needs real-device audition.