File size: 11,903 Bytes
b2e4883 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 | # 03 β Data Model
## Overview
PatternTalk's core data is small and human-editable. Most of it lives in JSON files under `apps/web/data/` and `data/training/`.
```
Prompt (user voice/text)
β
βΌ
ParsedRequest βββ onomatopoeia.json + prompt grammar
β
βΌ
Pattern (loaded from patterns/*.json, expanded to bars + tempo)
β
βΌ
MIDI events
β
βΌ
.mid file
```
## Prompt grammar
User input is freeform text, but parses into a structured `ParsedRequest`.
```typescript
interface ParsedRequest {
// Source pattern
onomatopoeia?: string; // e.g. "tupatupatupa"
patternId?: string; // e.g. "d-beat", "blast-traditional"
patternName?: string; // e.g. "D-Beat", user-friendly
// Length
bars: number; // default 4
beats?: number; // override pattern beats per bar
// Tempo
tempo?: number; // explicit BPM
tempoSource: "prompt" | "reaper" | "audio" | "default";
tempoDefault: number; // 120 if nothing else
// Time signature
timeSignature: {
numerator: number; // default 4
denominator: number; // default 4
};
// Cymbal/voice hints
cymbal?: {
type: "crash" | "ride" | "china" | "hihat-open" | "hihat-closed" | "splash";
pattern?: "8ths" | "quarters" | "bell" | "wash" | "upstrokes";
};
// Accent overrides
accents?: ("1" | "and-of-2" | "3" | "and-of-4" | string)[];
// Style modifiers
feel?: "straight" | "swing" | "half-time" | "double-time";
intensity?: "soft" | "medium" | "brutal" | "brutal-max";
}
```
## Onomatopoeia mapping
`apps/web/data/onomatopoeia.json` is the lookup table the parser checks first.
```json
{
"version": 1,
"entries": [
{
"id": "skank-beat",
"patterns": ["tupatupatupa", "tupa-tupa-tupa", "chka-chka-chka"],
"patternId": "skank",
"defaultCymbal": { "type": "hihat-open", "pattern": "upstrokes" },
"confidence": 0.9,
"notes": "Ska-style offbeat hi-hat upstrokes."
},
{
"id": "blast-beat",
"patterns": ["krrk-krrk-krrk", "BLAM-BLAM-BLAM", "krkrkrkrk"],
"patternId": "blast-traditional",
"defaultIntensity": "brutal",
"confidence": 0.85,
"notes": "Traditional blast beat. Kick-snare alternation."
},
{
"id": "boom-bap",
"patterns": ["boom-bap", "boom-bap-boom-bap"],
"patternId": "hiphop-basic",
"confidence": 0.95
},
{
"id": "china-accent",
"patterns": ["tss", "chka", "BLAM"],
"patternId": "china-accent",
"defaultCymbal": { "type": "china", "pattern": "wash" }
},
{
"id": "ride-bell",
"patterns": ["ding-ding-ding", "ding-ding-ding-ding"],
"patternId": "ride-bell-8ths",
"defaultCymbal": { "type": "ride", "pattern": "bell" }
}
]
}
```
### Matching logic
1. Normalize input: lowercase, strip punctuation, collapse whitespace
2. Phonetic match: Soundex or simple Levenshtein against entries
3. Syllable count: "tupatupatupa" = 4 syllables β likely 8th-note hi-hat
4. Confidence threshold: 0.7 to auto-map, lower to ask user
### User-defined onomatopoeias
Users can add their own mappings via the UI. Stored in `localStorage` per device.
```typescript
interface UserOnomatopoeia {
phrase: string;
patternId: string;
cymbal?: CymbalHint;
createdAt: number;
}
```
## Pattern templates
`apps/web/data/patterns/*.json` β one file per pattern. Hand-authored. **This is your moat.**
```typescript
interface PatternTemplate {
id: string; // unique slug
name: string; // "D-Beat"
description: string; // screenreader-friendly text
tags: string[]; // searchable
genre: "metal" | "rock" | "punk" | "pop" | "jazz" | "funk" | "latin" | "hiphop";
defaultTempo: number; // BPM
defaultBars: number; // how many bars this template covers
swingRatio?: number; // 0.0 = straight, 0.33 = shuffle
hits: Hit[];
}
interface Hit {
position: number; // 0.0 to 1.0, fraction of one bar
limb: Limb;
velocity: number; // 0-127 MIDI velocity
accent?: boolean; // visual + audio emphasis
}
type Limb =
| "kick" // right foot
| "snare" // left hand (default) or right hand (cross-handed)
| "hihat" // right hand (default) or left hand (cross-handed)
| "hihat-open" // same as hihat but with open hi-hat sample
| "ride"
| "ride-bell"
| "crash"
| "china"
| "splash"
| "tom-1" // highest
| "tom-2"
| "tom-3" // lowest floor tom
| "cross-stick"; // rim click
```
### Example: D-Beat
`apps/web/data/patterns/d-beat.json`:
```json
{
"id": "d-beat",
"name": "D-Beat",
"description": "D-beat pattern at 180 BPM. Kick plays on every beat with 8th-note doubles, snare hits on 2 and 4, ride bell plays steady 8ths throughout. Common in Discharge, Entombed, and Scandinavian crust punk.",
"tags": ["metal", "punk", "discharge", "entombed", "crust", "fast"],
"genre": "metal",
"defaultTempo": 180,
"defaultBars": 1,
"swingRatio": 0,
"hits": [
{ "position": 0.0, "limb": "kick", "velocity": 110 },
{ "position": 0.125, "limb": "kick", "velocity": 80 },
{ "position": 0.25, "limb": "snare", "velocity": 100 },
{ "position": 0.375, "limb": "kick", "velocity": 80 },
{ "position": 0.5, "limb": "kick", "velocity": 110 },
{ "position": 0.625, "limb": "kick", "velocity": 80 },
{ "position": 0.75, "limb": "snare", "velocity": 100 },
{ "position": 0.875, "limb": "kick", "velocity": 80 },
{ "position": 0.0, "limb": "ride-bell", "velocity": 70 },
{ "position": 0.125, "limb": "ride-bell", "velocity": 70 },
{ "position": 0.25, "limb": "ride-bell", "velocity": 70 },
{ "position": 0.375, "limb": "ride-bell", "velocity": 70 },
{ "position": 0.5, "limb": "ride-bell", "velocity": 70 },
{ "position": 0.625, "limb": "ride-bell", "velocity": 70 },
{ "position": 0.75, "limb": "ride-bell", "velocity": 70 },
{ "position": 0.875, "limb": "ride-bell", "velocity": 70 }
]
}
```
### Pattern library β starting list
**Metal (priority, 8 patterns):**
- `d-beat.json` β D-beat
- `blast-traditional.json` β traditional blast (kick-snare)
- `blast-hammer.json` β hammer-and-tongs blast
- `blast-hyper.json` β hyperblast (double kick 16ths)
- `half-time-metal.json` β Mastodon/Baroness half-time
- `djent-polyrhythm.json` β Meshuggah-style 4-over-3 kick
- `doom-slow.json` β Sleep/Electric Wizard doom
- `groove-metal.json` β mid-tempo groove metal
**Punk/Hardcore (4 patterns):**
- `punk-rock.json` β Ramones/Misfits 4-on-floor
- `hardcore.json` β Black Flag/Converge
- `skank.json` β ska skank beat (hi-hat upstrokes)
- `reggae-one-drop.json` β reggae
**Rock/Pop (4 patterns):**
- `rock-basic.json` β AC/DC/Beatles standard rock
- `rock-half-time.json` β half-time shuffle
- `pop-groove.json` β modern pop
- `country-train.json` β train beat
**Other genres (4 patterns):**
- `bossa-nova.json` β bossa nova
- `jazz-swing.json` β jazz ride swing
- `funk-new-orleans.json` β NOLA funk
- `hiphop-basic.json` β boom-bap
**Target: 18β20 patterns before hackathon, more as time permits.**
## Pattern engine
Lives in `apps/web/lib/patterns/engine.ts`.
### Inputs
- `PatternTemplate`
- `bars: number`
- `tempoBpm: number`
- `timeSignature: { numerator, denominator }`
- Optional `cymbal` override (replaces the default cymbal)
- Optional `accents` (boosts velocity on specified positions)
- Optional `feel` modifier
### Process
```
1. Load template
2. Expand to requested bars:
- If template.defaultBars === requested bars: use as-is
- Else: repeat the pattern N times, optionally vary velocity (humanize)
3. Apply cymbal override:
- If cymbal specified: replace ride/hihat hits with cymbal type
- Velocity scaled: cymbal typically quieter than snare
4. Apply accents:
- Find positions matching accent specs
- Boost velocity by +20
5. Apply feel modifier:
- "swing": delay 8th notes on the "and" by swingRatio
- "half-time": halve the effective tempo of snare/hihat
- "double-time": double it
6. Convert positions (0.0-1.0) to absolute ticks at tempoBpm
7. Return array of MIDI events
```
### Output
```typescript
interface MidiEvent {
tick: number; // absolute tick position
limb: Limb;
velocity: number; // 0-127
duration: number; // ticks (typically 1-10 for drums)
}
```
## MIDI generation
`apps/web/lib/midi/generator.ts` uses `@tonejs/midi`.
```typescript
import { Midi } from "@tonejs/midi";
function eventsToMidi(events: MidiEvent[]): Uint8Array {
const midi = new Midi();
midi.header.setTempo(events.bpm); // bpm on master track
midi.header.timeSignature = [4, 4];
const drums = midi.addTrack();
drums.name = "PatternTalk Drums";
for (const ev of events) {
drums.addNote({
midi: limbToGeneralMidi(ev.limb),
ticks: ev.tick,
durationTicks: ev.duration,
velocity: ev.velocity / 127,
});
}
return midi.toArray();
}
function limbToGeneralMidi(limb: Limb): number {
// GM Drum Map
const map: Record<Limb, number> = {
"kick": 36,
"snare": 38,
"cross-stick": 37,
"hihat": 42,
"hihat-open": 46,
"ride": 51,
"ride-bell": 53,
"crash": 49,
"china": 52,
"splash": 55,
"tom-1": 50,
"tom-2": 47,
"tom-3": 45,
};
return map[limb];
}
```
## Audio sample prompts
When PatternTalk calls the audio service, the prompt includes both the pattern context and a style hint:
```typescript
interface AudioPrompt {
pattern: string; // pattern name
limb?: Limb; // specific sample to generate
styleHints: string[]; // ["brutal drums", "18-inch china", "trashy"]
durationSeconds: number; // 1-5 for one-shots, 5-15 for loops
intensity: "soft" | "medium" | "brutal" | "brutal-max";
loraAdapter?: string; // "brutal-drums" by default
}
```
Examples:
- Limb-specific one-shot: `{ pattern: "D-Beat", limb: "china", styleHints: ["trashy", "18-inch", "panic-attack"], durationSeconds: 3, intensity: "brutal-max" }`
- Loop preview: `{ pattern: "D-Beat", styleHints: ["brutal drums", "tight snare"], durationSeconds: 8, intensity: "brutal" }`
## Training data manifest
`data/training/manifest.yaml` describes every audio file used to train the brutal-drum LoRA.
```yaml
version: 1
lora:
name: patterntalk-brutal-drums
base_model: stable-audio-3-small
training_steps: 1500
learning_rate: 1e-4
rank: 32
samples:
- id: kick-001
path: oneshots/kick-tight.wav
category: kick
tags: [tight, clicky, triggered]
source: original-recording
license: CC-BY
duration_seconds: 1.2
- id: snare-discharge-001
path: oneshots/snare-trashy.wav
category: snare
tags: [trashy, snappy, crust]
source: freesound.org/user-x
license: CC-BY
duration_seconds: 0.8
- id: loop-dbeat-discharge-style
path: loops/dbeat-180.wav
category: loop
tags: [d-beat, 180bpm, discharge-style]
source: original-recording
license: CC-BY
duration_seconds: 4.0
bpm: 180
```
**License discipline:** All samples must be CC0, CC-BY, or original recordings. No unlicensed material. Documented in the manifest, surfaced in the model card.
## What we're explicitly not modeling (yet)
- User accounts (out of scope)
- Cloud-saved pattern libraries (localStorage only)
- Pattern remixing/chaining
- Time-signature patterns other than 4/4 (the parser supports it but templates are 4/4)
- Polymeter / metric modulation
- Genre mixing (one pattern per request, not hybrids)
These are all reasonable post-hackathon extensions.
|