musehub_notation.py
python
| 1 | """MuseHub Notation Service — MIDI-to-standard-notation conversion. |
| 2 | |
| 3 | Converts MIDI note data (as stored in Muse commits) into quantized, structured |
| 4 | notation data suitable for rendering as sheet music. The output is a typed |
| 5 | JSON payload consumed by the client-side SVG score renderer. |
| 6 | |
| 7 | Why this exists |
| 8 | --------------- |
| 9 | Musicians who read traditional notation need to visualize Muse compositions as |
| 10 | sheet music without exporting to MusicXML and opening a separate application. |
| 11 | This service bridges the gap by producing quantized ``NotationResult`` data that |
| 12 | the score page renders directly in the browser. |
| 13 | |
| 14 | Design decisions |
| 15 | ---------------- |
| 16 | - Server-side quantization only. The browser renderer is intentionally thin |
| 17 | it receives pre-computed beat-aligned note data and draws SVG, it does not |
| 18 | re-quantize or re-interpret pitch. |
| 19 | - Deterministic stubs keyed on ``ref``. Full Storpheus MIDI introspection will |
| 20 | be wired in once the per-commit MIDI endpoint is stable. Until then, stub |
| 21 | data is musically realistic and internally consistent. |
| 22 | - No external I/O. This module is pure data — no database, no network calls, |
| 23 | no side effects. Route handlers inject all inputs. |
| 24 | |
| 25 | Boundary rules |
| 26 | -------------- |
| 27 | - Must NOT import StateStore, EntityRegistry, or executor modules. |
| 28 | - Must NOT import LLM handlers or muse_* pipeline modules. |
| 29 | - Must NOT import Storpheus service directly (data flows via route params). |
| 30 | """ |
| 31 | from __future__ import annotations |
| 32 | |
| 33 | |
| 34 | import hashlib |
| 35 | import logging |
| 36 | from typing import NamedTuple, TypedDict |
| 37 | |
| 38 | logger = logging.getLogger(__name__) |
| 39 | |
| 40 | # --------------------------------------------------------------------------- |
| 41 | # Public types |
| 42 | # --------------------------------------------------------------------------- |
| 43 | |
| 44 | |
| 45 | class NotationNote(TypedDict): |
| 46 | """A single quantized note ready for SVG rendering. |
| 47 | |
| 48 | Fields are kept flat (no nested objects) to simplify JavaScript |
| 49 | destructuring on the client side. |
| 50 | |
| 51 | pitch_name: e.g. "C", "F#", "Bb" |
| 52 | octave: MIDI octave number (4 = middle octave) |
| 53 | duration: note duration as a fraction string, e.g. "1/4", "1/8", "1/2" |
| 54 | start_beat: beat position (0-indexed from the start of the piece) |
| 55 | velocity: MIDI velocity 0–127 |
| 56 | track_id: source track index (matches NotationTrack.track_id) |
| 57 | """ |
| 58 | |
| 59 | pitch_name: str |
| 60 | octave: int |
| 61 | duration: str |
| 62 | start_beat: float |
| 63 | velocity: int |
| 64 | track_id: int |
| 65 | |
| 66 | |
| 67 | class NotationTrack(TypedDict): |
| 68 | """One instrument part, with clef/key/time signature metadata.""" |
| 69 | |
| 70 | track_id: int |
| 71 | clef: str |
| 72 | key_signature: str |
| 73 | time_signature: str |
| 74 | instrument: str |
| 75 | notes: list[NotationNote] |
| 76 | |
| 77 | |
| 78 | class NotationResult(NamedTuple): |
| 79 | """Typed result returned by ``convert_ref_to_notation``. |
| 80 | |
| 81 | Attributes |
| 82 | ---------- |
| 83 | tracks: |
| 84 | List of ``NotationTrack`` dicts, one per instrument part. Each track |
| 85 | includes clef, key_signature, time_signature, and a list of |
| 86 | ``NotationNote`` dicts ordered by start_beat. |
| 87 | tempo: |
| 88 | BPM as a positive integer. |
| 89 | key: |
| 90 | Key signature string, e.g. ``"C major"``, ``"F# minor"``. |
| 91 | time_sig: |
| 92 | Time signature string, e.g. ``"4/4"``, ``"3/4"``, ``"6/8"``. |
| 93 | """ |
| 94 | |
| 95 | tracks: list[NotationTrack] |
| 96 | tempo: int |
| 97 | key: str |
| 98 | time_sig: str |
| 99 | |
| 100 | |
| 101 | # --------------------------------------------------------------------------- |
| 102 | # Internal helpers |
| 103 | # --------------------------------------------------------------------------- |
| 104 | |
| 105 | _PITCH_NAMES = ["C", "C#", "D", "D#", "E", "F", "F#", "G", "G#", "A", "A#", "B"] |
| 106 | |
| 107 | _KEY_POOL = [ |
| 108 | "C major", |
| 109 | "G major", |
| 110 | "D major", |
| 111 | "A major", |
| 112 | "F major", |
| 113 | "Bb major", |
| 114 | "Eb major", |
| 115 | "A minor", |
| 116 | "D minor", |
| 117 | "E minor", |
| 118 | "B minor", |
| 119 | "G minor", |
| 120 | ] |
| 121 | |
| 122 | _CLEF_MAP = { |
| 123 | "bass": "bass", |
| 124 | "piano": "treble", |
| 125 | "keys": "treble", |
| 126 | "guitar": "treble", |
| 127 | "strings": "treble", |
| 128 | "violin": "treble", |
| 129 | "cello": "bass", |
| 130 | "trumpet": "treble", |
| 131 | "sax": "treble", |
| 132 | "default": "treble", |
| 133 | } |
| 134 | |
| 135 | _TIME_SIGS = ["4/4", "3/4", "6/8", "2/4"] |
| 136 | _DURATIONS = ["1/4", "1/4", "1/4", "1/8", "1/8", "1/2", "1/4", "1/4"] |
| 137 | _ROLE_NAMES = ["piano", "bass", "guitar", "strings", "trumpet"] |
| 138 | |
| 139 | |
| 140 | def _seed_from_ref(ref: str) -> int: |
| 141 | """Derive a deterministic integer seed from a commit ref string.""" |
| 142 | digest = hashlib.sha256(ref.encode()).digest() |
| 143 | return int.from_bytes(digest[:4], "big") |
| 144 | |
| 145 | |
| 146 | def _lcg(seed: int) -> int: |
| 147 | """Minimal linear congruential generator step — returns updated state.""" |
| 148 | return (seed * 1664525 + 1013904223) & 0xFFFFFFFF |
| 149 | |
| 150 | |
| 151 | def _notes_for_track( |
| 152 | seed: int, |
| 153 | track_idx: int, |
| 154 | time_sig: str, |
| 155 | num_bars: int, |
| 156 | ) -> list[NotationNote]: |
| 157 | """Generate a list of quantized notation notes for one track. |
| 158 | |
| 159 | Uses a seeded pseudo-random sequence so that the same ref always produces |
| 160 | the same notes. The quantization grid matches the time signature — quarter |
| 161 | notes for 4/4 and 3/4, eighth notes for 6/8. |
| 162 | """ |
| 163 | beats_per_bar, _ = (int(x) for x in time_sig.split("/")) |
| 164 | notes: list[NotationNote] = [] |
| 165 | |
| 166 | s = seed ^ (track_idx * 0xDEAD) |
| 167 | for bar in range(num_bars): |
| 168 | beat = 0.0 |
| 169 | while beat < beats_per_bar: |
| 170 | s = _lcg(s) |
| 171 | # 30 % chance of a rest — skip this beat slot |
| 172 | if (s % 10) < 3: |
| 173 | beat += 1.0 |
| 174 | continue |
| 175 | s = _lcg(s) |
| 176 | pitch_idx = s % 12 |
| 177 | s = _lcg(s) |
| 178 | octave = 3 + (s % 3) # octaves 3, 4, 5 |
| 179 | s = _lcg(s) |
| 180 | dur_idx = s % len(_DURATIONS) |
| 181 | duration = _DURATIONS[dur_idx] |
| 182 | s = _lcg(s) |
| 183 | velocity = 60 + (s % 60) |
| 184 | |
| 185 | notes.append( |
| 186 | NotationNote( |
| 187 | pitch_name=_PITCH_NAMES[pitch_idx], |
| 188 | octave=int(octave), |
| 189 | duration=duration, |
| 190 | start_beat=float(bar * beats_per_bar + beat), |
| 191 | velocity=int(velocity), |
| 192 | track_id=track_idx, |
| 193 | ) |
| 194 | ) |
| 195 | # Advance beat by the duration value (quarter = 1, eighth = 0.5, half = 2) |
| 196 | num, denom = (int(x) for x in duration.split("/")) |
| 197 | beat_advance = 4.0 * num / denom |
| 198 | beat += beat_advance |
| 199 | |
| 200 | return notes |
| 201 | |
| 202 | |
| 203 | # --------------------------------------------------------------------------- |
| 204 | # Public API |
| 205 | # --------------------------------------------------------------------------- |
| 206 | |
| 207 | |
| 208 | def convert_ref_to_notation( |
| 209 | ref: str, |
| 210 | num_tracks: int = 3, |
| 211 | num_bars: int = 8, |
| 212 | ) -> NotationResult: |
| 213 | """Convert a Muse commit ref to quantized notation data. |
| 214 | |
| 215 | Returns a ``NotationResult`` containing typed track data ready for the |
| 216 | client-side SVG score renderer. |
| 217 | |
| 218 | Parameters |
| 219 | ---------- |
| 220 | ref: |
| 221 | Muse commit ref (branch name, tag, or commit SHA). Used as a seed so |
| 222 | that the same ref always returns the same notation. |
| 223 | num_tracks: |
| 224 | Number of instrument tracks to generate. Clamped to [1, 8]. |
| 225 | num_bars: |
| 226 | Number of bars of music to generate per track. Clamped to [1, 32]. |
| 227 | """ |
| 228 | num_tracks = max(1, min(8, num_tracks)) |
| 229 | num_bars = max(1, min(32, num_bars)) |
| 230 | |
| 231 | seed = _seed_from_ref(ref) |
| 232 | |
| 233 | key_idx = seed % len(_KEY_POOL) |
| 234 | key = _KEY_POOL[key_idx] |
| 235 | |
| 236 | ts_seed = _lcg(seed) |
| 237 | time_sig = _TIME_SIGS[ts_seed % len(_TIME_SIGS)] |
| 238 | |
| 239 | tempo_seed = _lcg(ts_seed) |
| 240 | tempo = 80 + int(tempo_seed % 80) |
| 241 | |
| 242 | tracks: list[NotationTrack] = [] |
| 243 | for i in range(num_tracks): |
| 244 | role = _ROLE_NAMES[i % len(_ROLE_NAMES)] |
| 245 | clef = _CLEF_MAP.get(role, _CLEF_MAP["default"]) |
| 246 | notes = _notes_for_track(seed, i, time_sig, num_bars) |
| 247 | tracks.append( |
| 248 | NotationTrack( |
| 249 | track_id=i, |
| 250 | clef=clef, |
| 251 | key_signature=key, |
| 252 | time_signature=time_sig, |
| 253 | instrument=role, |
| 254 | notes=notes, |
| 255 | ) |
| 256 | ) |
| 257 | |
| 258 | logger.debug("✅ notation: ref=%s tracks=%d bars=%d tempo=%d", ref, num_tracks, num_bars, tempo) |
| 259 | return NotationResult(tracks=tracks, tempo=tempo, key=key, time_sig=time_sig) |
| 260 | |
| 261 | |
| 262 | class NotationDict(TypedDict): |
| 263 | """Serialized form of ``NotationResult`` for JSON API responses.""" |
| 264 | |
| 265 | tracks: list[NotationTrack] |
| 266 | tempo: int |
| 267 | key: str |
| 268 | timeSig: str |
| 269 | |
| 270 | |
| 271 | def notation_result_to_dict(result: NotationResult) -> NotationDict: |
| 272 | """Serialise a ``NotationResult`` to a typed dict for JSON responses. |
| 273 | |
| 274 | The output shape is: |
| 275 | ```json |
| 276 | { |
| 277 | "tracks": [...], |
| 278 | "tempo": 120, |
| 279 | "key": "C major", |
| 280 | "timeSig": "4/4" |
| 281 | } |
| 282 | ``` |
| 283 | |
| 284 | Note: ``time_sig`` is camelCase ``timeSig`` in the JSON output to match |
| 285 | the JavaScript convention used by all other MuseHub API responses. |
| 286 | """ |
| 287 | return NotationDict( |
| 288 | tracks=result.tracks, |
| 289 | tempo=result.tempo, |
| 290 | key=result.key, |
| 291 | timeSig=result.time_sig, |
| 292 | ) |