strudel-sample-library

name: strudel-sample-library description: Complete reference for every sound our Strudel tracks can use — the loaded sample banks (VCSL orchestral mallets/winds, ~50 drum kits, mridangam, the full TidalCycles Dirt set, eddyflux/crate world percussion, amen breaks), the 128 General MIDI soundfont instruments (gm_violin, gm_cello, gm_string_ensemble, gm_flute, gm_choir_aahs, gm_pad_*, gm_lead_*…), the built-in synths + FM, and how to load more (github: packs, shabda→Freesound, the open-strudel-samples explorer). Use when picking a voice, reaching for instrument variety, or when a sound isn't playing.

What samples we can use, what they sound like, and how to add more.

Currently loaded by default in our player

The boot() loader in web/src/engine/strudel.ts calls samples() + registerSoundfonts() on ALL of these at startup. Everything here is available with no extra loading:

So in practice you already have piano, the whole VCSL mallet/wind set, mridangam/SP-12/Linn/808/909 drums, the full TidalCycles Dirt library, organic world percussion, the amen breaks, AND the 128 General MIDI instruments with no extra loading. Real sustained strings finally exist — gm_string_ensemble_1, gm_cello, gm_synth_strings_1 — so you no longer have to fake a string pad with detuned sawtooths (though that's still a fine Kiasmos/Rone move). Rule of thumb: VCSL for plucked/mallet color (kalimba, marimba, vibraphone); GM for bowed/blown/sustained (strings, flute, sax, organ, choir); synths (saw/square/tri/sine + fm) for electronic.

Drum kits inside tidal-drum-machines you reach with .bank("..."):

AkaiLinn (warm/brushy, our default) · AkaiMPC60 · RolandTR808 · RolandTR909 · KorgKR55 · KorgKPR77 · OberheimDMX · RolandCR8000 · RolandTR606 · BossDR55 · RolandTR707 · ~50 more

Each kit has the same vocabulary: bd cp cb cr hh ht lt mt oh rd sd (+ kit-specific extras like sh shaker, tb tambourine).

.bank() is required — s("bd") alone silently drops. Always chain .bank("AkaiLinn").

General MIDI instruments (gm_*) — the full 128

Registered at boot via @strudel/soundfonts. Play like any pitched voice: note("C4 Eb4 G4").s("gm_string_ensemble_1"). The soundfont data lazy-loads on first use (a brief one-time delay per gm_ voice). This is the orchestra + band + GM synths — reach here for real bowed/blown/sustained voices the synths and VCSL can't give. Every option:

⚠️ gm_ soundfonts render SILENT in the offline WAV render (tools/render-wav.mjs). Each note's await getFontBufferSource(...) (async buffer decode) loses the race against OfflineAudioContext.startRendering(), so the buffers are empty. A gm_-only section measures ALL_ZERO; a mixed section measures only its synths + samples. Consequences: (a) the render→measure "ears loop" is BLIND to every gm_ voice — a gm_-heavy track can read TOO_SOFT purely because its melodic body is invisible (do NOT crank gains to "fix" this; you'll clip the live mix); (b) past gm_-heavy cranks' "pure-tone" metrics (aeon, lumen, glade…) actually measured only their synth/sample voices. For offline-measurable tracks, compose with synths (sawtooth/triangle/sine/square/noise) + samples (drums/piano/VCSL) — that's why HOME's [[crank-halo]] went all-synth. For gm_-based styles (orchestral/chamber), verify by EAR in the live player (localhost:5273) — gm_ loads fine in realtime; the offline render is just a harshness/level check on the synth/sample layer. (Fixable in render.ts by pre-decoding each voice's buffers before startRendering() — not yet done.)

• Piano / keys: gm_piano gm_acoustic_piano gm_bright_acoustic_piano gm_electric_grand_piano gm_honky_tonk_piano gm_epiano1 gm_epiano2 gm_harpsichord gm_clavinet
• Chromatic / mallet: gm_celesta gm_glockenspiel gm_music_box gm_vibraphone gm_marimba gm_xylophone gm_tubular_bells gm_dulcimer gm_tinkle_bell
• Organ: gm_drawbar_organ gm_percussive_organ gm_rock_organ gm_church_organ gm_reed_organ gm_accordion gm_bandoneon gm_harmonica
• Guitar: gm_acoustic_guitar_nylon gm_acoustic_guitar_steel gm_electric_guitar_clean gm_electric_guitar_jazz gm_electric_guitar_muted gm_overdriven_guitar gm_distortion_guitar gm_guitar_harmonics gm_guitar_fret_noise gm_banjo
• Bass: gm_acoustic_bass gm_electric_bass_finger gm_electric_bass_pick gm_fretless_bass gm_slap_bass_1 gm_slap_bass_2 gm_synth_bass_1 gm_synth_bass_2
• Strings: gm_violin gm_viola gm_cello gm_contrabass gm_fiddle gm_tremolo_strings gm_pizzicato_strings gm_orchestral_harp gm_string_ensemble_1 gm_string_ensemble_2 gm_synth_strings_1 gm_synth_strings_2
• Choir / voice: gm_choir_aahs gm_voice_oohs gm_synth_choir gm_orchestra_hit
• Brass: gm_trumpet gm_trombone gm_tuba gm_muted_trumpet gm_french_horn gm_brass_section gm_synth_brass_1 gm_synth_brass_2
• Reed / sax: gm_soprano_sax gm_alto_sax gm_tenor_sax gm_baritone_sax gm_oboe gm_english_horn gm_bassoon gm_clarinet
• Pipe / flute: gm_piccolo gm_flute gm_recorder gm_pan_flute gm_blown_bottle gm_shakuhachi gm_whistle gm_ocarina
• Synth lead: gm_lead_1_square gm_lead_2_sawtooth gm_lead_3_calliope gm_lead_4_chiff gm_lead_5_charang gm_lead_6_voice gm_lead_7_fifths gm_lead_8_bass_lead
• Synth pad: gm_pad_new_age gm_pad_warm gm_pad_poly gm_pad_choir gm_pad_bowed gm_pad_metallic gm_pad_halo gm_pad_sweep — ⚠️ these names DROP the GM program number. It's gm_pad_warm / gm_fx_crystal, not gm_pad_2_warm / gm_fx_3_crystal (the GM-numbered form throws sound … not found and renders SILENT). Same trap elsewhere: gm_acoustic_piano (not gm_acoustic_grand_piano), gm_epiano1/gm_epiano2 (not gm_electric_piano_1/2), gm_synth_choir (not gm_synth_voice), gm_accordion (not gm_tango_accordion). Synth leads are the exception — they DO keep their numbers (gm_lead_2_sawtooth). When unsure, grep the registered set: curl -s https://unpkg.com/@strudel/soundfonts@1.3.0/dist/index.mjs | grep -oE "gm_[a-z0-9_]+" | sort -u.
• Synth FX: gm_fx_rain gm_fx_soundtrack gm_fx_crystal gm_fx_atmosphere gm_fx_brightness gm_fx_goblins gm_fx_echoes gm_fx_sci_fi
• Ethnic: gm_sitar gm_shamisen gm_koto gm_kalimba gm_bagpipe gm_shanai gm_steel_drums
• Percussive: gm_agogo gm_woodblock gm_taiko_drum gm_melodic_tom gm_synth_drum gm_timpani gm_reverse_cymbal
• Sound FX: gm_breath_noise gm_seashore gm_bird_tweet gm_telephone gm_helicopter gm_applause gm_gunshot

Built-in synths + FM (no sample load)

• Oscillators: sawtooth square triangle sine (+ pulse, supersaw). Noise: white pink brown. Chip engine: zzfx.
• FM synthesis: any synth + .fm(N) (modulation index) + .fmh(N) (harmonic ratio) + .fmattack/.fmdecay — bells, metallic, growl. e.g. note("c2").s("sine").fm(4).fmh(2).
• Make these "different": .crush(N) (bit-crush), .coarse(N) (downsample), .shape(N)/.distort(N) (waveshape), .vowel("a e i o u") (formant), .phaser(N), plus full per-note .attack/.decay/.sustain/.release. The complete list lives in [[strudel-effects]] + [[strudel-modifiers]].

Mixing pitfalls that read as "not playing" (learned the hard way)

A voice can be in the stack(), parse fine, and STILL be inaudible. The usual culprits — check these before assuming a sound "doesn't work":

• Drum gain floor. Sample drums under ~0.18 gain vanish under synths. Hats/shakers want 0.25–0.5, not 0.12–0.18 (see [[strudel-conduct]] gotcha: "gain ranges that look fine sound silent").
• Don't over-hpf a hat. hpf(4200) strips almost all of a hi-hat's body and it goes silent. Keep shaker/hat hpf around 1200–2000.
• A low-passed kick disappears in a dense mix. s("bd").lpf(150) is a fine sparse-intro thud, but in a full section (kick + sub + pad all stacked low) it has no transient to cut through and reads as "not there." For groove/climax sections give the kick presence — lpf ≥ ~1000 or none — and keep sub/pad out of its low-end way.
• Same-family voices mask each other. A vibraphone counter under a vibraphone_bowed pad muddies — vary the timbre (use marimba/kalimba/harp for the counter).
• A pure sine sub is FELT, not heard. A low sine (e.g. C2 ≈ 65 Hz) puts all its energy at the fundamental, which laptop/phone speakers barely reproduce — so it seems missing no matter the gain. Give it harmonics that land in ~150–400 Hz: use triangle (warm) or sawtooth instead of sine and open the lpf to ~600–800, and/or octave-double (same note one octave up). Then the bass translates on any speaker.
• VCSL samples are recorded ~40× quieter than the piano/synths. Measured peak amplitudes: VCSL marimba/vibraphone ≈ 0.009, dough-samples piano ≈ 0.40. So a VCSL voice at gain(0.3–0.5) is buried ~40:1 under a piano or synth at the same gain — it plays, you just can't hear it (it'll still highlight in the editor, since that's pattern-driven, independent of audio level). If a VCSL sound highlights but is inaudible, it's almost certainly too quiet, NOT broken (not bit-depth, not decode — those are fine): give it gain(6–10), it has the headroom. Sustained/stacked VCSL (a pad) reads fine because levels accumulate; single short hits (marimba) vanish. Diagnosis trick: decode the WAV and read getChannelData(0) peak — if it's ~0.01 vs piano's ~0.4, that's your answer.
• Soundfont strings read LOW too. gm_string_ensemble_1 (and the other gm_ bowed/orchestral voices) output quietly — not 40:1 like VCSL, but enough that a string bed left at synth-pad gain (0.08–0.10) is basically inaudible. Push gm_ string .gain() to ~0.2–0.3+ (layering root/3rd/5th helps, but each voice still needs real level). They're warm AND present once given the gain — see the gyre/grotto string beds. Standing rule: reach for strings → start their gain high and trust it.

What's in the full TidalCycles Dirt set (loaded by default)

The github:tidalcycles/Dirt-Samples load gives 218 folders — far more than the dough-samples curated 9. Highlights you can reference right now (no loading needed):

(Already loaded by web/src/engine/strudel.ts via samples('github:tidalcycles/Dirt-Samples') — Strudel resolves github:user/repo → https://raw.githubusercontent.com/user/repo/main/strudel.json. Just reference the sound names directly.)

Available to add — not pre-loaded (on-demand)

Shabda → Freesound · the wildcard

Query Freesound.org by keyword directly inside samples(). No pack to install — fetches on demand, caches client-side:

await samples('shabda:bass:4,hihat:4')        // 4 basses + 4 hihats
await samples('shabda:water:8,thunder:4')     // 8 water sounds + 4 thunder
await samples('shabda:nes:8,arcade:4,coin:2') // chip-tune samples
await samples('shabda:zelda:4,mario_jump:2')  // specific game stems
await samples('shabda:vinyl_crackle:4')       // texture for Bonobo dustiness
await samples('shabda:owl:4,wind:4,fire:4')   // ambient field-recording
await samples('shabda:cello_pizz:8')          // pizzicato cello stems for Kiasmos
await samples('shabda/speech:dawn,light,wake') // synthetic TTS — say words as samples

Format: shabda:<keyword>:<count> joined with commas, or shabda/speech:word1,word2,... for TTS.

When to use:

• One-off texture you don't want to commit (don't load a whole pack for one wind sample)
• Specific real-world sounds Strudel doesn't ship (cello pizzicato, sea waves, footsteps)
• Voice samples — TTS "say" words as sounds, chop into a DJRUM-style stem palette

Caveats:

• Freesound search is unpredictable — shabda:violin:4 might return 4 wildly different violin recordings, not 4 takes of the same line
• Quality varies (community uploads). Re-roll if you don't like what you got
• Network fetch on first use (typically <500ms per sample)

Chip-tune / video-game sounds

There's no curated NES/SNES Strudel pack on GitHub, but three working approaches:

// 1. DIY chip-tune from synth primitives + crush/coarse
note("<C4 E4 G4 C5>").s("square").crush(4).coarse(3).gain(0.4)  // NES square
note("<G3>").s("triangle").crush(3).gain(0.5)                    // NES triangle bass
s("white").crush(2).hpf(4000).gain(0.15)                         // NES noise channel

// 2. Strudel's built-in ZZFX synth — tiny 8-bit-style engine
note("C4").s("zzfx").gain(0.5)

// 3. Shabda → Freesound for ad-hoc retro samples
await samples('shabda:nes_blip:4,coin:2,jump:2')

Free chip-tune sources you could convert to strudel.json manually (each is a folder of WAVs you'd host + write a manifest for):

• Woolyss · Chipmusic 8-bit / 16-bit soundfonts — Gameboy, NES/Famicom, SNES, Sega Genesis
• 8bitsamples.com — NES-style instruments/drums (some free, some paid)

Custom URLs

await samples({
  myKick:    'kick/01.wav',
  myShaker:  'perc/shaker.wav',
}, 'https://example.com/my-samples/');

Community packs (browse before installing)

• therebelrobot/open-strudel-samples — explorer UI for finding community-curated packs
• strudel-samples.alternet.site — search + preview every public strudel.json on GitHub before loading
• vasilymilovidov/samples — personal but high-quality

The awesome-strudel bank index (load any with github:<user>/<repo>)

Every public bank catalogued in awesome-strudel. We already load the three marked ✓ at boot; the rest are one await samples('github:…') away. Descriptions are given only where the name/contents are unambiguous — preview unfamiliar packs in the explorer above before committing (community uploads vary in quality and content).

(To make any of these load by default in our player, add await m.samples('github:…') to boot() in web/src/engine/strudel.ts — see "How to actually load more" above.)

Sample → artist style cheat-sheet

If you're writing for one of our reference artists, reach for these:

Bonobo (Black Sands)

• kalimba / kalimba2-5 (VCSL) — THE signature voice. Layer 2-3 variants at slightly different pans.
• marimba / vibraphone (VCSL) — supporting melodic
• AkaiLinn bd/sd — woody brushed kit
• flick / glasstap — sample-y accents

Skee Mask (Compro / Pool)

• RolandTR808 or RolandTR909 bd/sd — sharp electronic kit
• breaks125 / breaks152 (full Dirt) — jungle break under broken pattern
• wind / crow / insect — granular interlude bed
• hh*16 with degradeBy — Skee Mask spraying hat
• s("white").chop(16) — granular noise grain texture

Kiasmos

• vcsl orchestral strings (no direct "violin" sample, but layered sax + sustained sawtooth substitute)
• tubularbells for the pad-like bell ostinato
• RolandTR909 bd — pure 4×4 techno kick
• cello-substitute = layered detuned sawtooths

Floating Points (Crush / Cascade)

• mridangam hits — hand-percussion polyrhythms
• vibraphone / vibraphone_bowed — modal melody
• Buchla-substitute = sawtooth with perlin.range(-0.12, 0.12) detune
• RolandTR909 bd — club tracks

DJRUM

• piano + low-pass + room — contact-mic'd grit
• mridangam / tabla — hand percussion
• sax_vib — improvised solo voice
• breaks165 — the jungle break under ambient
• speakspell — chopped vocal stems

How to actually load more in our player

Add to the boot() loader in web/src/engine/strudel.ts (the banks array, the github: loads, or another m.samples(...) call):

await m.samples(`${SAMPLE_BASE}/vcsl.json`);     // a dough-samples pack
await m.samples('github:eddyflux/crate');        // any github repo with a strudel.json
await samples('shabda:cello:8,strings:8');       // Freesound on-demand (per-track works too)

The fetches are ~10–50KB JSON each (just manifests). Actual WAVs lazy-load on first reference. Don't load packs you won't use — they're free to add later.

Find more packs: the open-strudel-samples explorer and strudel-samples.alternet.site search every public strudel.json on GitHub — preview, then load with github:<user>/<repo>. Beyond what we load: eddyflux/crate, algorave-dave/samples, Bubobubobubobubo/Dough-Amen, Bubobubobubobubo/Dough-Juj.

Enumerate EVERYTHING at runtime

This skill lists the headline voices, but the live registry is ground truth. web/src/engine/strudel.ts exposes listSounds() — it returns every registered sound (all loaded sample banks + the 128 gm_* + the synths), which is what the on-screen keyboard / instrument picker shows. When you need the exhaustive instrument list, call it. For the exhaustive modifier / method list, [[strudel-modifiers]] catalogs every combinator and [[strudel-effects]] every filter/effect — together they are the complete method menu.

Sources

Strudel docs · Samples · felixroos/dough-samples · tidalcycles/Dirt-Samples · Open Strudel Samples explorer · vasilymilovidov/samples · awesome-strudel