Goonjan commited on
Commit
621ebcd
·
1 Parent(s): 29cf188

Finishing MVP

Browse files
README.md CHANGED
@@ -74,19 +74,17 @@ The quick test: can a pianist sight-read it at moderate tempo, clearly assign ha
74
  ## Roadmap
75
 
76
  **v1 — in progress**
77
- - [ ] End-to-end pipeline: audio → MIDI
78
- - [ ] Three core playability transforms (density, span, note cap)
79
- - [ ] Web UI with MIDI download
80
 
81
  **v2 — planned**
82
  - [ ] Chord-aware left hand voicing (root + third + fifth from chord analysis)
83
  - [ ] MuseScore PDF rendering
84
- - [ ] Before/after example gallery
85
 
86
  **v3 — planned**
87
  - [ ] Beat tracking with madmom for better metric strength scoring
88
  - [ ] Melody smoothing — strip ornaments and melisma from vocal transcription
89
- - [ ] Difficulty score on output
90
 
91
  **Later**
92
  - [ ] Fine-tuned arrangement model on POP909 dataset
 
74
  ## Roadmap
75
 
76
  **v1 — in progress**
77
+ - [x] End-to-end pipeline: audio → MIDI
78
+ - [x] Three core playability transforms (density, span, note cap)
79
+ - [x] Web UI with MIDI download
80
 
81
  **v2 — planned**
82
  - [ ] Chord-aware left hand voicing (root + third + fifth from chord analysis)
83
  - [ ] MuseScore PDF rendering
 
84
 
85
  **v3 — planned**
86
  - [ ] Beat tracking with madmom for better metric strength scoring
87
  - [ ] Melody smoothing — strip ornaments and melisma from vocal transcription
 
88
 
89
  **Later**
90
  - [ ] Fine-tuned arrangement model on POP909 dataset
src/keyarrange/piano/transforms.py CHANGED
@@ -22,7 +22,7 @@ def _group_by_onset(notes: list[Note], window_ms: float = 50.0) -> list[list[Not
22
 
23
  return groups
24
 
25
- def density_reducer(notes: list[Note], bpm: float) -> list[Note]:
26
  window_duration = 0.5 # 500ms
27
  step_size = 0.25 # 250ms
28
  dropped_note_ids = set()
@@ -37,7 +37,7 @@ def density_reducer(notes: list[Note], bpm: float) -> list[Note]:
37
  while t <= last_note_start_time + window_duration: # Extend windowing slightly past the last note's start time
38
  window_notes = [note for note in notes if t <= note.start < t + window_duration and note.id not in dropped_note_ids]
39
 
40
- max_notes_in_window = max(1, int(120 / bpm))
41
 
42
  if len(window_notes) > max_notes_in_window:
43
  # Sort by duration (longest first) and keep only the top `max_notes_in_window`
 
22
 
23
  return groups
24
 
25
+ def density_reducer(notes: list[Note], bpm: float, multiplier: int = 1) -> list[Note]:
26
  window_duration = 0.5 # 500ms
27
  step_size = 0.25 # 250ms
28
  dropped_note_ids = set()
 
37
  while t <= last_note_start_time + window_duration: # Extend windowing slightly past the last note's start time
38
  window_notes = [note for note in notes if t <= note.start < t + window_duration and note.id not in dropped_note_ids]
39
 
40
+ max_notes_in_window = max(1, int(120 / bpm)) * multiplier
41
 
42
  if len(window_notes) > max_notes_in_window:
43
  # Sort by duration (longest first) and keep only the top `max_notes_in_window`
src/keyarrange/pipeline.py CHANGED
@@ -61,9 +61,9 @@ class Pipeline:
61
  quantized_left_notes = quantize_to_beats(left_notes, beat_times)
62
 
63
  logger.info("Applying transformations to Right hand notes...")
64
- # right_notes = density_reducer(right_notes, bpm)
65
- # right_notes = span_enforcer(right_notes, max_span=12, hand="right")
66
- # right_notes = note_cap(right_notes, max_notes=3)
67
 
68
  logger.info("Applying transformations to Left hand notes...")
69
  left_notes = density_reducer(quantized_left_notes, bpm)
 
61
  quantized_left_notes = quantize_to_beats(left_notes, beat_times)
62
 
63
  logger.info("Applying transformations to Right hand notes...")
64
+ right_notes = density_reducer(right_notes, bpm, multiplier=2) # Allow density relaxation for vocals
65
+ right_notes = span_enforcer(right_notes, max_span=12, hand="right")
66
+ right_notes = note_cap(right_notes, max_notes=3)
67
 
68
  logger.info("Applying transformations to Left hand notes...")
69
  left_notes = density_reducer(quantized_left_notes, bpm)
web/index.html CHANGED
@@ -4,6 +4,10 @@
4
  <meta charset="UTF-8">
5
  <meta name="viewport" content="width=device-width, initial-scale=1.0">
6
  <title>KeyArrange</title>
 
 
 
 
7
  <style>
8
  @import url('https://fonts.googleapis.com/css2?family=DM+Serif+Display:ital@0;1&family=DM+Mono:wght@400;500&family=DM+Sans:wght@300;400;500&display=swap');
9
 
@@ -56,10 +60,7 @@
56
  margin-bottom: 16px;
57
  }
58
 
59
- h1 em {
60
- font-style: italic;
61
- color: var(--accent);
62
- }
63
 
64
  .subtitle {
65
  font-size: 14px;
@@ -68,10 +69,8 @@
68
  line-height: 1.7;
69
  }
70
 
71
- /* Upload section */
72
- .upload-section {
73
- margin-bottom: 32px;
74
- }
75
 
76
  .drop-zone {
77
  border: 1px dashed var(--border);
@@ -98,19 +97,8 @@
98
  height: 100%;
99
  }
100
 
101
- .drop-icon {
102
- font-family: 'DM Mono', monospace;
103
- font-size: 28px;
104
- margin-bottom: 12px;
105
- color: var(--text-dim);
106
- }
107
-
108
- .drop-label {
109
- font-size: 14px;
110
- color: var(--text-mid);
111
- margin-bottom: 6px;
112
- }
113
-
114
  .drop-sub {
115
  font-family: 'DM Mono', monospace;
116
  font-size: 11px;
@@ -129,7 +117,6 @@
129
  border: 1px solid rgba(200, 240, 67, 0.2);
130
  border-radius: 3px;
131
  }
132
-
133
  .file-selected.visible { display: flex; }
134
 
135
  .file-name {
@@ -151,10 +138,8 @@
151
  line-height: 1;
152
  padding: 0 4px;
153
  }
154
-
155
  .file-clear:hover { color: var(--text); }
156
 
157
- /* Arrange button */
158
  .btn-arrange {
159
  width: 100%;
160
  padding: 16px;
@@ -171,11 +156,7 @@
171
  margin-top: 12px;
172
  transition: all 0.15s;
173
  }
174
-
175
- .btn-arrange:hover:not(:disabled) {
176
- background: #d8ff55;
177
- }
178
-
179
  .btn-arrange:disabled {
180
  background: var(--border);
181
  color: var(--text-dim);
@@ -191,7 +172,6 @@
191
  border-radius: 3px;
192
  margin-top: 20px;
193
  }
194
-
195
  .status-card.visible { display: block; }
196
 
197
  .status-eyebrow {
@@ -203,11 +183,7 @@
203
  margin-bottom: 8px;
204
  }
205
 
206
- .status-message {
207
- font-size: 14px;
208
- color: var(--text-mid);
209
- line-height: 1.6;
210
- }
211
 
212
  .progress-bar-track {
213
  height: 2px;
@@ -221,8 +197,6 @@
221
  height: 100%;
222
  background: var(--accent3);
223
  border-radius: 2px;
224
- width: 0%;
225
- transition: width 0.6s ease;
226
  animation: indeterminate 2s ease-in-out infinite;
227
  }
228
 
@@ -244,15 +218,10 @@
244
  color: #ff6b6b;
245
  font-family: 'DM Mono', monospace;
246
  }
247
-
248
  .error-card.visible { display: block; }
249
 
250
  /* Results */
251
- .results-section {
252
- display: none;
253
- margin-top: 32px;
254
- }
255
-
256
  .results-section.visible { display: block; }
257
 
258
  .results-header {
@@ -281,8 +250,8 @@
281
  color: var(--text);
282
  }
283
 
284
- /* Piano roll */
285
- .piano-roll-wrap {
286
  background: var(--surface);
287
  border: 1px solid var(--border);
288
  border-radius: 3px;
@@ -290,48 +259,62 @@
290
  margin-bottom: 20px;
291
  }
292
 
293
- .piano-roll-label {
294
  font-family: 'DM Mono', monospace;
295
  font-size: 10px;
296
  letter-spacing: 0.15em;
297
  text-transform: uppercase;
298
  color: var(--text-dim);
299
- padding: 12px 16px 0;
300
- }
301
-
302
- .piano-roll-wrap img {
303
- width: 100%;
304
- display: block;
305
  }
306
 
307
- .piano-roll-legend {
308
  display: flex;
309
- gap: 20px;
310
- padding: 10px 16px 14px;
311
  }
312
 
313
  .legend-item {
314
  display: flex;
315
  align-items: center;
316
- gap: 7px;
317
- font-size: 12px;
318
- color: var(--text-mid);
319
  }
320
 
321
  .legend-dot {
322
- width: 10px;
323
- height: 10px;
324
  border-radius: 2px;
325
  flex-shrink: 0;
326
  }
327
 
328
- /* Download */
329
- .download-row {
330
- display: flex;
331
- gap: 10px;
332
- flex-wrap: wrap;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
333
  }
334
 
 
 
 
335
  .btn-download {
336
  padding: 12px 20px;
337
  background: transparent;
@@ -349,12 +332,7 @@
349
  align-items: center;
350
  gap: 8px;
351
  }
352
-
353
- .btn-download:hover {
354
- border-color: var(--accent);
355
- color: var(--accent);
356
- }
357
-
358
  .btn-download .btn-icon { font-size: 15px; }
359
 
360
  /* Footer */
@@ -370,7 +348,6 @@
370
  flex-wrap: wrap;
371
  gap: 8px;
372
  }
373
-
374
  footer a { color: var(--text-dim); text-decoration: none; }
375
  footer a:hover { color: var(--accent); }
376
  </style>
@@ -404,11 +381,11 @@
404
  <!-- Status -->
405
  <div class="status-card" id="statusCard">
406
  <div class="status-eyebrow">Processing</div>
407
- <div class="status-message" id="statusMessage">
408
  Separating stems, transcribing, arranging… this takes 1–3 minutes depending on song length.
409
  </div>
410
  <div class="progress-bar-track">
411
- <div class="progress-bar-fill" id="progressFill"></div>
412
  </div>
413
  </div>
414
 
@@ -422,22 +399,29 @@
422
  <h2 class="results-title">Your arrangement</h2>
423
  </div>
424
 
425
- <div class="piano-roll-wrap" id="pianoRollWrap" style="display:none">
426
- <div class="piano-roll-label">Piano Roll · time → pitch ↑</div>
427
- <img id="pianoRollImg" src="" alt="Piano roll visualization" />
428
- <div class="piano-roll-legend">
429
- <div class="legend-item">
430
- <div class="legend-dot" style="background:#4a9eff"></div>
431
- Right hand — melody
432
- </div>
433
- <div class="legend-item">
434
- <div class="legend-dot" style="background:#ff6b6b"></div>
435
- Left hand — harmony
436
- </div>
437
- <div class="legend-item" style="margin-left:auto; font-size:11px; color:var(--text-dim)">
438
- Dashed line = middle C
439
  </div>
440
  </div>
 
 
 
 
 
 
 
 
 
441
  </div>
442
 
443
  <div class="download-row">
@@ -464,16 +448,13 @@
464
  const fileClear = document.getElementById('fileClear');
465
  const arrangeBtn = document.getElementById('arrangeBtn');
466
  const statusCard = document.getElementById('statusCard');
467
- const statusMsg = document.getElementById('statusMessage');
468
  const errorCard = document.getElementById('errorCard');
469
  const results = document.getElementById('resultsSection');
470
- const pianoWrap = document.getElementById('pianoRollWrap');
471
- const pianoImg = document.getElementById('pianoRollImg');
472
  const midiDownload = document.getElementById('midiDownload');
473
 
474
  let selectedFile = null;
475
 
476
- // Drag-over styling
477
  dropZone.addEventListener('dragover', e => { e.preventDefault(); dropZone.classList.add('drag-over'); });
478
  dropZone.addEventListener('dragleave', () => dropZone.classList.remove('drag-over'));
479
  dropZone.addEventListener('drop', e => {
@@ -518,22 +499,16 @@
518
  const res = await fetch('/upload', { method: 'POST', body: form });
519
  const data = await res.json();
520
 
521
- if (!res.ok) {
522
- throw new Error(data.detail || 'Something went wrong.');
523
- }
524
 
525
  hide(statusCard);
526
  show(results);
527
 
 
 
 
528
  midiDownload.href = data.midi_url;
529
 
530
- if (data.piano_roll_url) {
531
- pianoImg.src = data.piano_roll_url;
532
- pianoWrap.style.display = 'block';
533
- } else {
534
- pianoWrap.style.display = 'none';
535
- }
536
-
537
  } catch (err) {
538
  hide(statusCard);
539
  errorCard.textContent = '⚠ ' + err.message;
 
4
  <meta charset="UTF-8">
5
  <meta name="viewport" content="width=device-width, initial-scale=1.0">
6
  <title>KeyArrange</title>
7
+
8
+ <!-- MIDI player web component — handles audio synthesis + scrolling piano roll visualization -->
9
+ <script src="https://cdn.jsdelivr.net/combine/npm/tone@14.7.58,npm/@magenta/music@1.23.1/es6/core.js,npm/html-midi-player@1.5.0"></script>
10
+
11
  <style>
12
  @import url('https://fonts.googleapis.com/css2?family=DM+Serif+Display:ital@0;1&family=DM+Mono:wght@400;500&family=DM+Sans:wght@300;400;500&display=swap');
13
 
 
60
  margin-bottom: 16px;
61
  }
62
 
63
+ h1 em { font-style: italic; color: var(--accent); }
 
 
 
64
 
65
  .subtitle {
66
  font-size: 14px;
 
69
  line-height: 1.7;
70
  }
71
 
72
+ /* Upload */
73
+ .upload-section { margin-bottom: 32px; }
 
 
74
 
75
  .drop-zone {
76
  border: 1px dashed var(--border);
 
97
  height: 100%;
98
  }
99
 
100
+ .drop-icon { font-size: 28px; margin-bottom: 12px; color: var(--text-dim); }
101
+ .drop-label { font-size: 14px; color: var(--text-mid); margin-bottom: 6px; }
 
 
 
 
 
 
 
 
 
 
 
102
  .drop-sub {
103
  font-family: 'DM Mono', monospace;
104
  font-size: 11px;
 
117
  border: 1px solid rgba(200, 240, 67, 0.2);
118
  border-radius: 3px;
119
  }
 
120
  .file-selected.visible { display: flex; }
121
 
122
  .file-name {
 
138
  line-height: 1;
139
  padding: 0 4px;
140
  }
 
141
  .file-clear:hover { color: var(--text); }
142
 
 
143
  .btn-arrange {
144
  width: 100%;
145
  padding: 16px;
 
156
  margin-top: 12px;
157
  transition: all 0.15s;
158
  }
159
+ .btn-arrange:hover:not(:disabled) { background: #d8ff55; }
 
 
 
 
160
  .btn-arrange:disabled {
161
  background: var(--border);
162
  color: var(--text-dim);
 
172
  border-radius: 3px;
173
  margin-top: 20px;
174
  }
 
175
  .status-card.visible { display: block; }
176
 
177
  .status-eyebrow {
 
183
  margin-bottom: 8px;
184
  }
185
 
186
+ .status-message { font-size: 14px; color: var(--text-mid); line-height: 1.6; }
 
 
 
 
187
 
188
  .progress-bar-track {
189
  height: 2px;
 
197
  height: 100%;
198
  background: var(--accent3);
199
  border-radius: 2px;
 
 
200
  animation: indeterminate 2s ease-in-out infinite;
201
  }
202
 
 
218
  color: #ff6b6b;
219
  font-family: 'DM Mono', monospace;
220
  }
 
221
  .error-card.visible { display: block; }
222
 
223
  /* Results */
224
+ .results-section { display: none; margin-top: 32px; }
 
 
 
 
225
  .results-section.visible { display: block; }
226
 
227
  .results-header {
 
250
  color: var(--text);
251
  }
252
 
253
+ /* MIDI player container */
254
+ .player-wrap {
255
  background: var(--surface);
256
  border: 1px solid var(--border);
257
  border-radius: 3px;
 
259
  margin-bottom: 20px;
260
  }
261
 
262
+ .player-label {
263
  font-family: 'DM Mono', monospace;
264
  font-size: 10px;
265
  letter-spacing: 0.15em;
266
  text-transform: uppercase;
267
  color: var(--text-dim);
268
+ padding: 12px 16px 8px;
269
+ display: flex;
270
+ align-items: center;
271
+ justify-content: space-between;
 
 
272
  }
273
 
274
+ .player-legend {
275
  display: flex;
276
+ gap: 16px;
 
277
  }
278
 
279
  .legend-item {
280
  display: flex;
281
  align-items: center;
282
+ gap: 6px;
283
+ font-size: 11px;
284
+ color: var(--text-dim);
285
  }
286
 
287
  .legend-dot {
288
+ width: 9px;
289
+ height: 9px;
290
  border-radius: 2px;
291
  flex-shrink: 0;
292
  }
293
 
294
+ /*
295
+ html-midi-player uses a shadow DOM so deep CSS overrides are limited.
296
+ These CSS custom properties are what the component officially exposes.
297
+ The visualizer background and note colors are the most impactful ones.
298
+ */
299
+ midi-player {
300
+ display: block;
301
+ width: 100%;
302
+ background: var(--surface2);
303
+ border-top: 1px solid var(--border);
304
+ }
305
+
306
+ midi-visualizer {
307
+ display: block;
308
+ width: 100%;
309
+ /* Note colors per track index — track 0 = right hand, track 1 = left hand */
310
+ --midi-visualizer-notes-color: #4a9eff; /* fallback / right hand */
311
+ background: #0a0a0a;
312
+ min-height: 160px;
313
  }
314
 
315
+ /* Download */
316
+ .download-row { display: flex; gap: 10px; flex-wrap: wrap; }
317
+
318
  .btn-download {
319
  padding: 12px 20px;
320
  background: transparent;
 
332
  align-items: center;
333
  gap: 8px;
334
  }
335
+ .btn-download:hover { border-color: var(--accent); color: var(--accent); }
 
 
 
 
 
336
  .btn-download .btn-icon { font-size: 15px; }
337
 
338
  /* Footer */
 
348
  flex-wrap: wrap;
349
  gap: 8px;
350
  }
 
351
  footer a { color: var(--text-dim); text-decoration: none; }
352
  footer a:hover { color: var(--accent); }
353
  </style>
 
381
  <!-- Status -->
382
  <div class="status-card" id="statusCard">
383
  <div class="status-eyebrow">Processing</div>
384
+ <div class="status-message">
385
  Separating stems, transcribing, arranging… this takes 1–3 minutes depending on song length.
386
  </div>
387
  <div class="progress-bar-track">
388
+ <div class="progress-bar-fill"></div>
389
  </div>
390
  </div>
391
 
 
399
  <h2 class="results-title">Your arrangement</h2>
400
  </div>
401
 
402
+ <div class="player-wrap">
403
+ <div class="player-label">
404
+ <span>Piano Roll · press play to hear it</span>
405
+ <div class="player-legend">
406
+ <div class="legend-item">
407
+ <div class="legend-dot" style="background:#ff6b6b"></div>
408
+ Playing now
409
+ </div>
410
+ <div class="legend-item">
411
+ <div class="legend-dot" style="background:#4a9eff"></div>
412
+ Upcoming
413
+ </div>
 
 
414
  </div>
415
  </div>
416
+
417
+ <!--
418
+ midi-visualizer renders the scrolling piano roll.
419
+ midi-player handles play/pause/seek and audio via Magenta soundfont.
420
+ They are linked by the visualizer="#midiVisualizer" attribute.
421
+ sound-font enables the Magenta piano soundfont (requires internet).
422
+ -->
423
+ <midi-visualizer type="piano-roll" id="midiVisualizer"></midi-visualizer>
424
+ <midi-player id="midiPlayer" sound-font visualizer="#midiVisualizer"></midi-player>
425
  </div>
426
 
427
  <div class="download-row">
 
448
  const fileClear = document.getElementById('fileClear');
449
  const arrangeBtn = document.getElementById('arrangeBtn');
450
  const statusCard = document.getElementById('statusCard');
 
451
  const errorCard = document.getElementById('errorCard');
452
  const results = document.getElementById('resultsSection');
453
+ const midiPlayer = document.getElementById('midiPlayer');
 
454
  const midiDownload = document.getElementById('midiDownload');
455
 
456
  let selectedFile = null;
457
 
 
458
  dropZone.addEventListener('dragover', e => { e.preventDefault(); dropZone.classList.add('drag-over'); });
459
  dropZone.addEventListener('dragleave', () => dropZone.classList.remove('drag-over'));
460
  dropZone.addEventListener('drop', e => {
 
499
  const res = await fetch('/upload', { method: 'POST', body: form });
500
  const data = await res.json();
501
 
502
+ if (!res.ok) throw new Error(data.detail || 'Something went wrong.');
 
 
503
 
504
  hide(statusCard);
505
  show(results);
506
 
507
+ // Setting src on midi-player triggers it to fetch the MIDI and load it
508
+ // into both the player controls and the linked visualizer automatically.
509
+ midiPlayer.setAttribute('src', data.midi_url);
510
  midiDownload.href = data.midi_url;
511
 
 
 
 
 
 
 
 
512
  } catch (err) {
513
  hide(statusCard);
514
  errorCard.textContent = '⚠ ' + err.message;