forked from jason9693/MusicTransformer-tensorflow2.0
-
Notifications
You must be signed in to change notification settings - Fork 0
/
sequence.py
861 lines (766 loc) · 30 KB
/
sequence.py
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
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
# import numpy as np
# import copy, itertools, collections
# from pretty_midi import PrettyMIDI, Note, Instrument
#
# # ==================================================================================
# # Parameters
# # ==================================================================================
#
# # NoteSeq -------------------------------------------------------------------------
#
# DEFAULT_SAVING_PROGRAM = 1
# DEFAULT_LOADING_PROGRAMS = range(128)
# DEFAULT_RESOLUTION = 220
# DEFAULT_TEMPO = 120
# DEFAULT_VELOCITY = 64
# # DEFAULT_PITCH_RANGE = range(21, 109)
# DEFAULT_PITCH_RANGE = range(0, 128)
# DEFAULT_VELOCITY_RANGE = range(21, 109)
# DEFAULT_NORMALIZATION_BASELINE = 60 # C4
#
# # EventSeq ------------------------------------------------------------------------
#
# USE_VELOCITY = True
# BEAT_LENGTH = 60 / DEFAULT_TEMPO
# # DEFAULT_TIME_SHIFT_BINS = 1.15 ** (np.arange(100) / 100)
# # DEFAULT_TIME_SHIFT_BINS = 1.15 ** np.arange(100) / (1.15 ** 99)
# DEFAULT_TIME_SHIFT_BINS = np.arange(1,101) / 100
#
#
# DEFAULT_VELOCITY_STEPS = 32
# DEFAULT_NOTE_LENGTH = BEAT_LENGTH * 2
# MIN_NOTE_LENGTH = BEAT_LENGTH / 2
#
# # ControlSeq ----------------------------------------------------------------------
#
# DEFAULT_WINDOW_SIZE = BEAT_LENGTH * 4
# DEFAULT_NOTE_DENSITY_BINS = np.arange(12) * 3 + 1
#
#
# # ==================================================================================
# # Notes
# # ==================================================================================
#
# class NoteSeq:
#
# @staticmethod
# def from_midi(midi, programs=DEFAULT_LOADING_PROGRAMS):
# notes = itertools.chain(*[
# inst.notes for inst in midi.instruments
# if inst.program in programs and not inst.is_drum])
# return NoteSeq(list(notes))
#
# @staticmethod
# def from_midi_file(path, *kargs, **kwargs):
# midi = PrettyMIDI(path)
# return NoteSeq.from_midi(midi, *kargs, **kwargs)
#
# @staticmethod
# def merge(*note_seqs):
# notes = itertools.chain(*[seq.notes for seq in note_seqs])
# return NoteSeq(list(notes))
#
# def __init__(self, notes=[]):
# self.notes = []
# if notes:
# for note in notes:
# assert isinstance(note, Note)
# notes = filter(lambda note: note.end >= note.start, notes)
# self.add_notes(list(notes))
#
# def copy(self):
# return copy.deepcopy(self)
#
# def to_midi(self, program=DEFAULT_SAVING_PROGRAM,
# resolution=DEFAULT_RESOLUTION, tempo=DEFAULT_TEMPO):
# midi = PrettyMIDI(resolution=resolution, initial_tempo=tempo)
# inst = Instrument(program, False, 'NoteSeq')
# inst.notes = copy.deepcopy(self.notes)
# midi.instruments.append(inst)
# return midi
#
# def to_midi_file(self, path, *kargs, **kwargs):
# self.to_midi(*kargs, **kwargs).write(path)
#
# def add_notes(self, notes):
# self.notes += notes
# self.notes.sort(key=lambda note: note.start)
#
# def adjust_pitches(self, offset):
# for note in self.notes:
# pitch = note.pitch + offset
# pitch = 0 if pitch < 0 else pitch
# pitch = 127 if pitch > 127 else pitch
# note.pitch = pitch
#
# def adjust_velocities(self, offset):
# for note in self.notes:
# velocity = note.velocity + offset
# velocity = 0 if velocity < 0 else velocity
# velocity = 127 if velocity > 127 else velocity
# note.velocity = velocity
#
# def adjust_time(self, offset):
# for note in self.notes:
# note.start += offset
# note.end += offset
#
# def trim_overlapped_notes(self, min_interval=0):
# last_notes = {}
# for i, note in enumerate(self.notes):
# if note.pitch in last_notes:
# last_note = last_notes[note.pitch]
# if note.start - last_note.start <= min_interval:
# last_note.end = max(note.end, last_note.end)
# last_note.velocity = max(note.velocity, last_note.velocity)
# del self.notes[i]
# elif note.start < last_note.end:
# last_note.end = note.start
# else:
# last_notes[note.pitch] = note
#
#
# # ==================================================================================
# # Events
# # ==================================================================================
#
# class Event:
#
# def __init__(self, type, time, value):
# self.type = type
# self.time = time
# self.value = value
#
# def __repr__(self):
# return 'Event(type={}, time={}, value={})'.format(
# self.type, self.time, self.value)
#
#
# class EventSeq:
# pitch_range = DEFAULT_PITCH_RANGE
# velocity_range = DEFAULT_VELOCITY_RANGE
# velocity_steps = DEFAULT_VELOCITY_STEPS
# time_shift_bins = DEFAULT_TIME_SHIFT_BINS
#
# @staticmethod
# def from_note_seq(note_seq):
# note_events = []
#
# if USE_VELOCITY:
# velocity_bins = EventSeq.get_velocity_bins()
#
# for note in note_seq.notes:
# if note.pitch in EventSeq.pitch_range:
# if USE_VELOCITY:
# velocity = note.velocity
# velocity = max(velocity, EventSeq.velocity_range.start)
# velocity = min(velocity, EventSeq.velocity_range.stop - 1)
# velocity_index = np.searchsorted(velocity_bins, velocity)
# note_events.append(Event('velocity', note.start, velocity_index))
#
# pitch_index = note.pitch - EventSeq.pitch_range.start
# note_events.append(Event('note_on', note.start, pitch_index))
# note_events.append(Event('note_off', note.end, pitch_index))
#
# note_events.sort(key=lambda event: event.time) # stable
# events = []
# sphere = 0
#
# for i, event in enumerate(note_events):
# events.append(event)
#
# if event is note_events[-1]:
# break
#
# interval = note_events[i + 1].time - event.time
# shift = 0
#
# while interval - shift >= EventSeq.time_shift_bins[0]:
# index = np.searchsorted(EventSeq.time_shift_bins,
# interval - shift, side='right') - 1
# events.append(Event('time_shift', event.time + shift - sphere, index))
# shift += EventSeq.time_shift_bins[index]
# # note_events[i+1].time -= (interval-shift)
# sphere += (interval-shift)
#
# return EventSeq(events)
#
# @staticmethod
# def from_array(event_indeces):
# time = 0
# events = []
# for event_index in event_indeces:
# for event_type, feat_range in EventSeq.feat_ranges().items():
# if feat_range.start <= event_index < feat_range.stop:
# event_value = event_index - feat_range.start
# events.append(Event(event_type, time, event_value))
# if event_type == 'time_shift':
# time += EventSeq.time_shift_bins[event_value]
# break
#
# return EventSeq(events)
#
# @staticmethod
# def dim():
# return sum(EventSeq.feat_dims().values())
#
# @staticmethod
# def feat_dims():
# feat_dims = collections.OrderedDict()
# feat_dims['time_shift'] = len(EventSeq.time_shift_bins)
# feat_dims['note_on'] = len(EventSeq.pitch_range)
# feat_dims['note_off'] = len(EventSeq.pitch_range)
# if USE_VELOCITY:
# feat_dims['velocity'] = EventSeq.velocity_steps
# return feat_dims
#
# @staticmethod
# def feat_ranges():
# offset = 0
# feat_ranges = collections.OrderedDict()
# for feat_name, feat_dim in EventSeq.feat_dims().items():
# feat_ranges[feat_name] = range(offset, offset + feat_dim)
# offset += feat_dim
# return feat_ranges
#
# @staticmethod
# def get_velocity_bins():
# n = EventSeq.velocity_range.stop - EventSeq.velocity_range.start
# return np.arange(
# EventSeq.velocity_range.start,
# EventSeq.velocity_range.stop,
# n / (EventSeq.velocity_steps - 1))
#
# def __init__(self, events=[]):
# for event in events:
# assert isinstance(event, Event)
#
# self.events = copy.deepcopy(events)
#
# # compute event times again
# time = 0
# for event in self.events:
# event.time = time
# if event.type == 'time_shift':
# time += EventSeq.time_shift_bins[event.value]
#
# def to_note_seq(self):
# time = 0
# notes = []
#
# velocity = DEFAULT_VELOCITY
# velocity_bins = EventSeq.get_velocity_bins()
#
# last_notes = {}
#
# for event in self.events:
# if event.type == 'note_on':
# pitch = event.value + EventSeq.pitch_range.start
# note = Note(velocity, pitch, time, None)
# notes.append(note)
# last_notes[pitch] = note
#
# elif event.type == 'note_off':
# pitch = event.value + EventSeq.pitch_range.start
#
# if pitch in last_notes:
# note = last_notes[pitch]
# note.end = max(time, note.start + MIN_NOTE_LENGTH)
# del last_notes[pitch]
#
# elif event.type == 'velocity':
# index = min(event.value, velocity_bins.size - 1)
# velocity = velocity_bins[index]
#
# elif event.type == 'time_shift':
# time += EventSeq.time_shift_bins[event.value]
#
# for note in notes:
# if note.end is None:
# note.end = note.start + DEFAULT_NOTE_LENGTH
#
# note.velocity = int(note.velocity)
#
# return NoteSeq(notes)
#
# def to_array(self):
# feat_idxs = EventSeq.feat_ranges()
# idxs = [feat_idxs[event.type][event.value] for event in self.events]
# dtype = np.uint8 if EventSeq.dim() <= 256 else np.uint16
# return np.array(idxs, dtype=dtype)
#
#
# # ==================================================================================
# # Controls
# # ==================================================================================
#
# class Control:
#
# def __init__(self, pitch_histogram, note_density):
# self.pitch_histogram = pitch_histogram # list
# self.note_density = note_density # int
#
# def __repr__(self):
# return 'Control(pitch_histogram={}, note_density={})'.format(
# self.pitch_histogram, self.note_density)
#
# def to_array(self):
# feat_dims = ControlSeq.feat_dims()
# ndens = np.zeros([feat_dims['note_density']])
# ndens[self.note_density] = 1. # [dens_dim]
# phist = np.array(self.pitch_histogram) # [hist_dim]
# return np.concatenate([ndens, phist], 0) # [dens_dim + hist_dim]
#
#
# class ControlSeq:
# note_density_bins = DEFAULT_NOTE_DENSITY_BINS
# window_size = DEFAULT_WINDOW_SIZE
#
# @staticmethod
# def from_event_seq(event_seq):
# events = list(event_seq.events)
# start, end = 0, 0
#
# pitch_count = np.zeros([12])
# note_count = 0
#
# controls = []
#
# def _rel_pitch(pitch):
# return (pitch - 24) % 12
#
# for i, event in enumerate(events):
#
# while start < i:
# if events[start].type == 'note_on':
# abs_pitch = events[start].value + EventSeq.pitch_range.start
# rel_pitch = _rel_pitch(abs_pitch)
# pitch_count[rel_pitch] -= 1.
# note_count -= 1.
# start += 1
#
# while end < len(events):
# if events[end].time - event.time > ControlSeq.window_size:
# break
# if events[end].type == 'note_on':
# abs_pitch = events[end].value + EventSeq.pitch_range.start
# rel_pitch = _rel_pitch(abs_pitch)
# pitch_count[rel_pitch] += 1.
# note_count += 1.
# end += 1
#
# pitch_histogram = (
# pitch_count / note_count
# if note_count
# else np.ones([12]) / 12
# ).tolist()
#
# note_density = max(np.searchsorted(
# ControlSeq.note_density_bins,
# note_count, side='right') - 1, 0)
#
# controls.append(Control(pitch_histogram, note_density))
#
# return ControlSeq(controls)
#
# @staticmethod
# def dim():
# return sum(ControlSeq.feat_dims().values())
#
# @staticmethod
# def feat_dims():
# note_density_dim = len(ControlSeq.note_density_bins)
# return collections.OrderedDict([
# ('pitch_histogram', 12),
# ('note_density', note_density_dim)
# ])
#
# @staticmethod
# def feat_ranges():
# offset = 0
# feat_ranges = collections.OrderedDict()
# for feat_name, feat_dim in ControlSeq.feat_dims().items():
# feat_ranges[feat_name] = range(offset, offset + feat_dim)
# offset += feat_dim
# return feat_ranges
#
# @staticmethod
# def recover_compressed_array(array):
# feat_dims = ControlSeq.feat_dims()
# assert array.shape[1] == 1 + feat_dims['pitch_histogram']
# ndens = np.zeros([array.shape[0], feat_dims['note_density']])
# ndens[np.arange(array.shape[0]), array[:, 0]] = 1. # [steps, dens_dim]
# phist = array[:, 1:].astype(np.float64) / 255 # [steps, hist_dim]
# return np.concatenate([ndens, phist], 1) # [steps, dens_dim + hist_dim]
#
# def __init__(self, controls):
# for control in controls:
# assert isinstance(control, Control)
# self.controls = copy.deepcopy(controls)
#
# def to_compressed_array(self):
# ndens = [control.note_density for control in self.controls]
# ndens = np.array(ndens, dtype=np.uint8).reshape(-1, 1)
# phist = [control.pitch_histogram for control in self.controls]
# phist = (np.array(phist) * 255).astype(np.uint8)
# return np.concatenate([
# ndens, # [steps, 1] density index
# phist # [steps, hist_dim] 0-255
# ], 1) # [steps, hist_dim + 1]
#
#
# if __name__ == '__main__':
# import pickle, sys
#
# path = sys.argv[1] if len(sys.argv) > 1 else 'dataset/midi/balamb.mid'
#
# print(EventSeq.dim())
#
# print('Converting MIDI to EventSeq')
# es = EventSeq.from_note_seq(NoteSeq.from_midi_file(path))
#
# print(NoteSeq.from_midi_file(path).notes)
# # print(NoteSeq.from_midi(NoteSeq.from_midi_file(path).to_midi()).notes)
# # assert NoteSeq.from_midi_file(path) == NoteSeq.from_midi(NoteSeq.from_midi_file(path).to_midi())
# print('Converting EventSeq to MIDI')
# print([(i, item) for i, item in enumerate(NoteSeq.from_midi_file(path).notes)])
# print([(i, item) for i, item in enumerate(EventSeq.from_array(es.to_array()).to_note_seq().notes)])
# #assert NoteSeq.from_midi_file(path).notes == EventSeq.from_array(es.to_array()).to_note_seq().notes
# assert (es.to_array() == EventSeq.from_array(es.to_array()).to_array()).all()
#
# mid = EventSeq.from_array(es.to_array()).to_note_seq().to_midi()
# print(NoteSeq.from_midi(mid).notes)
# EventSeq.from_array(es.to_array()).to_note_seq().to_midi_file('test.mid')
import numpy as np
import copy, itertools, collections
from pretty_midi import PrettyMIDI, Note, Instrument
# ==================================================================================
# Parameters
# ==================================================================================
# NoteSeq -------------------------------------------------------------------------
DEFAULT_SAVING_PROGRAM = 1
DEFAULT_LOADING_PROGRAMS = range(128)
DEFAULT_RESOLUTION = 220
DEFAULT_TEMPO = 120
DEFAULT_VELOCITY = 64
DEFAULT_PITCH_RANGE = range(21, 109) # 109-20 = 89
DEFAULT_VELOCITY_RANGE = range(21, 109) # 109 - 20 = 89
DEFAULT_NORMALIZATION_BASELINE = 60 # C4
# EventSeq ------------------------------------------------------------------------
USE_VELOCITY = True
BEAT_LENGTH = 60 / DEFAULT_TEMPO
DEFAULT_TIME_SHIFT_BINS = 1.15 ** np.arange(32) / 65
DEFAULT_VELOCITY_STEPS = 32
DEFAULT_NOTE_LENGTH = BEAT_LENGTH * 2
MIN_NOTE_LENGTH = BEAT_LENGTH / 2
# ControlSeq ----------------------------------------------------------------------
DEFAULT_WINDOW_SIZE = BEAT_LENGTH * 4
DEFAULT_NOTE_DENSITY_BINS = np.arange(12) * 3 + 1
# ==================================================================================
# Notes
# ==================================================================================
class NoteSeq:
@staticmethod
def from_midi(midi, programs=DEFAULT_LOADING_PROGRAMS):
notes = itertools.chain(*[
inst.notes for inst in midi.instruments
if inst.program in programs and not inst.is_drum])
return NoteSeq(list(notes))
@staticmethod
def from_midi_file(path, *kargs, **kwargs):
midi = PrettyMIDI(path)
return NoteSeq.from_midi(midi, *kargs, **kwargs)
@staticmethod
def merge(*note_seqs):
notes = itertools.chain(*[seq.notes for seq in note_seqs])
return NoteSeq(list(notes))
def __init__(self, notes=[]):
self.notes = []
if notes:
for note in notes:
assert isinstance(note, Note)
notes = filter(lambda note: note.end >= note.start, notes)
self.add_notes(list(notes))
def copy(self):
return copy.deepcopy(self)
def to_midi(self, program=DEFAULT_SAVING_PROGRAM,
resolution=DEFAULT_RESOLUTION, tempo=DEFAULT_TEMPO):
midi = PrettyMIDI(resolution=resolution, initial_tempo=tempo)
inst = Instrument(program, False, 'NoteSeq')
inst.notes = copy.deepcopy(self.notes)
midi.instruments.append(inst)
return midi
def to_midi_file(self, path, *kargs, **kwargs):
self.to_midi(*kargs, **kwargs).write(path)
def add_notes(self, notes):
self.notes += notes
self.notes.sort(key=lambda note: note.start)
def adjust_pitches(self, offset):
for note in self.notes:
pitch = note.pitch + offset
pitch = 0 if pitch < 0 else pitch
pitch = 127 if pitch > 127 else pitch
note.pitch = pitch
def adjust_velocities(self, offset):
for note in self.notes:
velocity = note.velocity + offset
velocity = 0 if velocity < 0 else velocity
velocity = 127 if velocity > 127 else velocity
note.velocity = velocity
def adjust_time(self, offset):
for note in self.notes:
note.start += offset
note.end += offset
def trim_overlapped_notes(self, min_interval=0):
last_notes = {}
for i, note in enumerate(self.notes):
if note.pitch in last_notes:
last_note = last_notes[note.pitch]
if note.start - last_note.start <= min_interval:
last_note.end = max(note.end, last_note.end)
last_note.velocity = max(note.velocity, last_note.velocity)
del self.notes[i]
elif note.start < last_note.end:
last_note.end = note.start
else:
last_notes[note.pitch] = note
# ==================================================================================
# Events
# ==================================================================================
class Event:
def __init__(self, type, time, value):
self.type = type
self.time = time
self.value = value
def __repr__(self):
return 'Event(type={}, time={}, value={})'.format(
self.type, self.time, self.value)
class EventSeq:
pitch_range = DEFAULT_PITCH_RANGE
velocity_range = DEFAULT_VELOCITY_RANGE
velocity_steps = DEFAULT_VELOCITY_STEPS
time_shift_bins = DEFAULT_TIME_SHIFT_BINS
@staticmethod
def from_note_seq(note_seq):
note_events = []
if USE_VELOCITY:
velocity_bins = EventSeq.get_velocity_bins()
for note in note_seq.notes:
if note.pitch in EventSeq.pitch_range:
if USE_VELOCITY:
velocity = note.velocity
velocity = max(velocity, EventSeq.velocity_range.start)
velocity = min(velocity, EventSeq.velocity_range.stop - 1)
velocity_index = np.searchsorted(velocity_bins, velocity)
note_events.append(Event('velocity', note.start, velocity_index))
pitch_index = note.pitch - EventSeq.pitch_range.start
note_events.append(Event('note_on', note.start, pitch_index))
note_events.append(Event('note_off', note.end, pitch_index))
note_events.sort(key=lambda event: event.time) # stable
events = []
for i, event in enumerate(note_events):
events.append(event)
if event is note_events[-1]:
break
interval = note_events[i + 1].time - event.time
shift = 0
while interval - shift >= EventSeq.time_shift_bins[0]:
index = np.searchsorted(EventSeq.time_shift_bins,
interval - shift, side='right') - 1
events.append(Event('time_shift', event.time + shift, index))
shift += EventSeq.time_shift_bins[index]
return EventSeq(events)
@staticmethod
def from_array(event_indeces):
time = 0
events = []
for event_index in event_indeces:
for event_type, feat_range in EventSeq.feat_ranges().items():
if feat_range.start <= event_index < feat_range.stop:
event_value = event_index - feat_range.start
events.append(Event(event_type, time, event_value))
if event_type == 'time_shift':
time += EventSeq.time_shift_bins[event_value]
break
return EventSeq(events)
@staticmethod
def dim():
return sum(EventSeq.feat_dims().values())
@staticmethod
def feat_dims():
feat_dims = collections.OrderedDict()
feat_dims['note_on'] = len(EventSeq.pitch_range)
feat_dims['note_off'] = len(EventSeq.pitch_range)
if USE_VELOCITY:
feat_dims['velocity'] = EventSeq.velocity_steps
feat_dims['time_shift'] = len(EventSeq.time_shift_bins)
return feat_dims
@staticmethod
def feat_ranges():
offset = 0
feat_ranges = collections.OrderedDict()
for feat_name, feat_dim in EventSeq.feat_dims().items():
feat_ranges[feat_name] = range(offset, offset + feat_dim)
offset += feat_dim
return feat_ranges
@staticmethod
def get_velocity_bins():
n = EventSeq.velocity_range.stop - EventSeq.velocity_range.start
return np.arange(
EventSeq.velocity_range.start,
EventSeq.velocity_range.stop,
n / (EventSeq.velocity_steps - 1))
def __init__(self, events=[]):
for event in events:
assert isinstance(event, Event)
self.events = copy.deepcopy(events)
# compute event times again
time = 0
for event in self.events:
event.time = time
if event.type == 'time_shift':
time += EventSeq.time_shift_bins[event.value]
def to_note_seq(self):
time = 0
notes = []
velocity = DEFAULT_VELOCITY
velocity_bins = EventSeq.get_velocity_bins()
last_notes = {}
for event in self.events:
if event.type == 'note_on':
pitch = event.value + EventSeq.pitch_range.start
note = Note(velocity, pitch, time, None)
notes.append(note)
last_notes[pitch] = note
elif event.type == 'note_off':
pitch = event.value + EventSeq.pitch_range.start
if pitch in last_notes:
note = last_notes[pitch]
note.end = max(time, note.start + MIN_NOTE_LENGTH)
del last_notes[pitch]
elif event.type == 'velocity':
index = min(event.value, velocity_bins.size - 1)
velocity = velocity_bins[index]
elif event.type == 'time_shift':
time += EventSeq.time_shift_bins[event.value]
for note in notes:
if note.end is None:
note.end = note.start + DEFAULT_NOTE_LENGTH
note.velocity = int(note.velocity)
return NoteSeq(notes)
def to_array(self):
feat_idxs = EventSeq.feat_ranges()
idxs = [feat_idxs[event.type][event.value] for event in self.events]
dtype = np.uint8 if EventSeq.dim() <= 256 else np.uint16
return np.array(idxs, dtype=dtype)
# ==================================================================================
# Controls
# ==================================================================================
class Control:
def __init__(self, pitch_histogram, note_density):
self.pitch_histogram = pitch_histogram # list
self.note_density = note_density # int
def __repr__(self):
return 'Control(pitch_histogram={}, note_density={})'.format(
self.pitch_histogram, self.note_density)
def to_array(self):
feat_dims = ControlSeq.feat_dims()
ndens = np.zeros([feat_dims['note_density']])
ndens[self.note_density] = 1. # [dens_dim]
phist = np.array(self.pitch_histogram) # [hist_dim]
return np.concatenate([ndens, phist], 0) # [dens_dim + hist_dim]
class ControlSeq:
note_density_bins = DEFAULT_NOTE_DENSITY_BINS
window_size = DEFAULT_WINDOW_SIZE
@staticmethod
def from_event_seq(event_seq):
events = list(event_seq.events)
start, end = 0, 0
pitch_count = np.zeros([12])
note_count = 0
controls = []
def _rel_pitch(pitch):
return (pitch - 24) % 12
for i, event in enumerate(events):
while start < i:
if events[start].type == 'note_on':
abs_pitch = events[start].value + EventSeq.pitch_range.start
rel_pitch = _rel_pitch(abs_pitch)
pitch_count[rel_pitch] -= 1.
note_count -= 1.
start += 1
while end < len(events):
if events[end].time - event.time > ControlSeq.window_size:
break
if events[end].type == 'note_on':
abs_pitch = events[end].value + EventSeq.pitch_range.start
rel_pitch = _rel_pitch(abs_pitch)
pitch_count[rel_pitch] += 1.
note_count += 1.
end += 1
pitch_histogram = (
pitch_count / note_count
if note_count
else np.ones([12]) / 12
).tolist()
note_density = max(np.searchsorted(
ControlSeq.note_density_bins,
note_count, side='right') - 1, 0)
controls.append(Control(pitch_histogram, note_density))
return ControlSeq(controls)
@staticmethod
def dim():
return sum(ControlSeq.feat_dims().values())
@staticmethod
def feat_dims():
note_density_dim = len(ControlSeq.note_density_bins)
return collections.OrderedDict([
('pitch_histogram', 12),
('note_density', note_density_dim)
])
@staticmethod
def feat_ranges():
offset = 0
feat_ranges = collections.OrderedDict()
for feat_name, feat_dim in ControlSeq.feat_dims().items():
feat_ranges[feat_name] = range(offset, offset + feat_dim)
offset += feat_dim
return feat_ranges
@staticmethod
def recover_compressed_array(array):
feat_dims = ControlSeq.feat_dims()
assert array.shape[1] == 1 + feat_dims['pitch_histogram']
ndens = np.zeros([array.shape[0], feat_dims['note_density']])
ndens[np.arange(array.shape[0]), array[:, 0]] = 1. # [steps, dens_dim]
phist = array[:, 1:].astype(np.float64) / 255 # [steps, hist_dim]
return np.concatenate([ndens, phist], 1) # [steps, dens_dim + hist_dim]
def __init__(self, controls):
for control in controls:
assert isinstance(control, Control)
self.controls = copy.deepcopy(controls)
def to_compressed_array(self):
ndens = [control.note_density for control in self.controls]
ndens = np.array(ndens, dtype=np.uint8).reshape(-1, 1)
phist = [control.pitch_histogram for control in self.controls]
phist = (np.array(phist) * 255).astype(np.uint8)
return np.concatenate([
ndens, # [steps, 1] density index
phist # [steps, hist_dim] 0-255
], 1) # [steps, hist_dim + 1]
if __name__ == '__main__':
import pickle, sys
path = sys.argv[1] if len(sys.argv) > 1 else 'dataset/sample/c_maj.mid'
print(EventSeq.dim())
print('Converting MIDI to EventSeq')
es = EventSeq.from_note_seq(NoteSeq.from_midi_file(path))
print('Converting EventSeq to MIDI')
EventSeq.from_array(es.to_array()[:30]).to_note_seq().to_midi_file('test.mid')
print(list(es.to_array()[:30]))
# print('Converting EventSeq to ControlSeq')
# cs = ControlSeq.from_event_seq(es)
# print('Saving compressed ControlSeq')
# pickle.dump(cs.to_compressed_array(), open('/tmp/cs-compressed.data', 'wb'))
#
# print('Loading compressed ControlSeq')
# c = ControlSeq.recover_compressed_array(pickle.load(open('/tmp/cs-compressed.data', 'rb')))
print('Done')