-
Notifications
You must be signed in to change notification settings - Fork 11
/
experiment_process_jingju_crnn.py
635 lines (543 loc) · 27.9 KB
/
experiment_process_jingju_crnn.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
# -*- coding: utf-8 -*-
import pickle
import os
import sys
import shutil
from os import makedirs
from os.path import isfile
from os.path import exists
import numpy as np
import pyximport
from madmom.features.onsets import OnsetPeakPickingProcessor
from eval_jingju import eval_write_2_txt
from experiment_process_helper import boundary_decoding
from experiment_process_helper import data_parser
from experiment_process_helper import get_boundary_list
from experiment_process_helper import get_line_properties
from experiment_process_helper import get_results_decoding_path
from experiment_process_helper import odf_calculation_crnn
from experiment_process_helper import write_results_2_txt_jingju
from plot_code import plot_jingju
sys.path.append(os.path.join(os.path.dirname(__file__), "./src/"))
sys.path.append(os.path.join(os.path.dirname(__file__), "./training_scripts/"))
from parameters_jingju import *
from file_path_jingju_shared import *
from labWriter import boundaryLabWriter
from trainTestSeparation import getTestRecordingsScoreDurCorrectionArtistAlbumFilter
from utilFunctions import smooth_obs
from training_scripts.models_CRNN import jan_original
from audio_preprocessing import getMFCCBands2DMadmom
pyximport.install(reload_support=True,
setup_args={'include_dirs': np.get_include()})
import viterbiDecoding
def batch_process_onset_detection(wav_path,
textgrid_path,
score_path,
scaler,
test_recordings,
model_keras_cnn_0,
cnnModel_name,
detection_results_path,
architecture,
len_seq,
lab=False,
threshold=0.54,
obs_cal=True,
decoding_method='viterbi',
stateful=True):
"""
experiment process, evaluate a whole jingju dataset using CRNN model
:param wav_path: string, where we store the wav
:param textgrid_path: string, where we store the textgrid
:param score_path: string, where we store the score
:param scaler: sklearn object, StandardScaler
:param test_recordings: list of strings, testing recording filename
:param model_keras_cnn_0: keras .h5, model weights
:param cnnModel_name: string, model name
:param detection_results_path: string, where we store the evaluation results
:param architecture: string, model architecture name
:param len_seq: input sequence frame length
:param lab: string, for Riyaz dataset, not used in the paper
:param threshold: float, threshold for peak picking onset selection
:param obs_cal: string, tocal or toload, for saving running time
:param decoding_method: string, viterbi or peakPicking
:param stateful: bool, whether to use the stateful RNN
:return:
"""
eval_results_decoding_path = \
get_results_decoding_path(decoding_method=decoding_method,
bool_corrected_score_duration=varin['corrected_score_duration'],
eval_results_path=detection_results_path)
for artist_path, rn in test_recordings:
score_file = join(score_path, artist_path, rn+'.csv')
if not isfile(score_file):
print 'Score not found: ' + score_file
continue
nested_syllable_lists, wav_file, line_list, syllables, syllable_durations, bpm, pinyins = \
data_parser(artist_path=artist_path,
wav_path=wav_path,
textgrid_path=textgrid_path,
rn=rn,
score_file=score_file,
lab=lab)
if obs_cal == 'tocal':
# load audio
mfcc = getMFCCBands2DMadmom(wav_file, fs, hopsize_t, channel=1)
mfcc_scaled = scaler.transform(mfcc)
i_line = -1
for i_obs, line in enumerate(line_list):
if not lab and len(line[2]) == 0:
continue
i_line += 1
try:
print(syllable_durations[i_line])
except IndexError:
continue
if float(bpm[i_line]) == 0:
continue
time_line, lyrics_line, frame_start, frame_end = get_line_properties(lab=lab,
line=line,
hopsize_t=hopsize_t)
obs_path = join('./obs', cnnModel_name, artist_path)
obs_filename = rn + '_' + str(i_line + 1) + '.pkl'
if obs_cal == 'tocal':
obs_i, mfcc_line = odf_calculation_crnn(mfcc=mfcc,
mfcc_scaled=mfcc_scaled,
model_keras_cnn_0=model_keras_cnn_0,
frame_start=frame_start,
frame_end=frame_end,
len_seq=len_seq,
stateful=stateful)
# save onset curve
print('save onset curve ... ...')
if not exists(obs_path):
makedirs(obs_path)
pickle.dump(obs_i, open(join(obs_path, obs_filename), 'w'))
else:
obs_i = pickle.load(open(join(obs_path, obs_filename), 'r'))
obs_i = np.squeeze(obs_i)
obs_i = smooth_obs(obs_i)
# organize score
print('Calculating: ', rn, ' phrase', str(i_obs))
print('ODF Methods: ', architecture)
duration_score = syllable_durations[i_line]
duration_score = np.array([float(ds) for ds in duration_score if len(ds)])
duration_score *= (time_line/np.sum(duration_score))
i_boundary, label = boundary_decoding(decoding_method=decoding_method,
obs_i=obs_i,
duration_score=duration_score,
varin=varin,
threshold=threshold,
hopsize_t=hopsize_t,
viterbiDecoding=viterbiDecoding,
OnsetPeakPickingProcessor=OnsetPeakPickingProcessor)
time_boundary_start = np.array(i_boundary[:-1])*hopsize_t
time_boundary_end = np.array(i_boundary[1:])*hopsize_t
boundary_list = get_boundary_list(lab=lab,
decoding_method=decoding_method,
time_boundary_start=time_boundary_start,
time_boundary_end=time_boundary_end,
pinyins=pinyins,
syllables=syllables,
i_line=i_line)
filename_syll_lab = join(eval_results_decoding_path,
artist_path, rn + '_' + str(i_line + 1) + '.syll.lab')
boundaryLabWriter(boundaryList=boundary_list,
outputFilename=filename_syll_lab,
label=label)
if varin['plot'] and obs_cal == 'tocal':
plot_jingju(nested_syllable_lists,
i_line,
mfcc_line,
hopsize_t,
obs_i,
i_boundary,
duration_score)
return eval_results_decoding_path
def viterbi_subroutine(test_nacta_2017,
test_nacta,
eval_label,
obs_cal,
len_seq,
model_name,
architecture,
scaler,
full_path_model,
detection_results_path):
"""routine for viterbi decoding"""
list_recall_onset_25, list_precision_onset_25, list_F1_onset_25 = [], [], []
list_recall_onset_5, list_precision_onset_5, list_F1_onset_5 = [], [], []
list_recall_25, list_precision_25, list_F1_25 = [], [], []
list_recall_5, list_precision_5, list_F1_5 = [], [], []
for ii in range(5):
if obs_cal == 'tocal':
stateful = False if varin['overlap'] else True
input_shape = (1, len_seq, 1, 80, 15)
# initialize the model
model_keras_cnn_0 = jan_original(filter_density=1,
dropout=0.5,
input_shape=input_shape,
batchNorm=False,
dense_activation='sigmoid',
channel=1,
stateful=stateful,
training=False,
bidi=varin['bidi'])
# load the model weights
model_keras_cnn_0.load_weights(full_path_model + str(ii) + '.h5')
# delete detection results path if it exists
detection_results_path_model = join(detection_results_path + str(ii))
if os.path.exists(detection_results_path_model) and os.path.isdir(detection_results_path + str(ii)):
shutil.rmtree(detection_results_path + str(ii))
if varin['dataset'] != 'ismir':
# evaluate nacta 2017 data set
batch_process_onset_detection(wav_path=nacta2017_wav_path,
textgrid_path=nacta2017_textgrid_path,
score_path=nacta2017_score_unified_path,
test_recordings=test_nacta_2017,
model_keras_cnn_0=model_keras_cnn_0,
len_seq=len_seq,
cnnModel_name=model_name + str(ii),
detection_results_path=detection_results_path + str(ii),
scaler=scaler,
obs_cal=obs_cal,
decoding_method='viterbi',
architecture=architecture,
stateful=stateful)
# evaluate nacta dataset
eval_results_decoding_path = \
batch_process_onset_detection(wav_path=nacta_wav_path,
textgrid_path=nacta_textgrid_path,
score_path=nacta_score_unified_path,
test_recordings=test_nacta,
model_keras_cnn_0=model_keras_cnn_0,
cnnModel_name=model_name + str(ii),
detection_results_path=detection_results_path + str(ii),
scaler=scaler,
obs_cal=obs_cal,
decoding_method='viterbi',
architecture=architecture,
stateful=stateful,
len_seq=len_seq)
else:
eval_results_decoding_path = detection_results_path + str(ii)
precision_onset, recall_onset, F1_onset, \
precision, recall, F1, \
= eval_write_2_txt(eval_result_file_name=join(eval_results_decoding_path, 'results.csv'),
segSyllable_path=eval_results_decoding_path,
label=eval_label,
decoding_method='viterbi')
list_precision_onset_25.append(precision_onset[0])
list_precision_onset_5.append(precision_onset[1])
list_recall_onset_25.append(recall_onset[0])
list_recall_onset_5.append(recall_onset[1])
list_F1_onset_25.append(F1_onset[0])
list_F1_onset_5.append(F1_onset[1])
list_precision_25.append(precision[0])
list_precision_5.append(precision[1])
list_recall_25.append(recall[0])
list_recall_5.append(recall[1])
list_F1_25.append(F1[0])
list_F1_5.append(F1[1])
return list_precision_onset_25, \
list_recall_onset_25, \
list_F1_onset_25, \
list_precision_25, \
list_recall_25, \
list_F1_25, \
list_precision_onset_5, \
list_recall_onset_5, \
list_F1_onset_5, \
list_precision_5, \
list_recall_5, \
list_F1_5
def peak_picking_subroutine(test_nacta_2017,
test_nacta,
th,
obs_cal,
scaler,
len_seq,
model_name,
full_path_model,
architecture,
detection_results_path,
jingju_eval_results_path):
"""routine for peak picking decoding"""
from src.utilFunctions import append_or_write
import csv
eval_result_file_name = join(jingju_eval_results_path,
varin['sample_weighting'],
model_name + '_peakPicking_threshold_results.txt')
list_recall_onset_25, list_precision_onset_25, list_F1_onset_25 = [], [], []
list_recall_onset_5, list_precision_onset_5, list_F1_onset_5 = [], [], []
list_recall_25, list_precision_25, list_F1_25 = [], [], []
list_recall_5, list_precision_5, list_F1_5 = [], [], []
for ii in range(5):
stateful = False if varin['overlap'] else True
if obs_cal == 'tocal':
input_shape = (1, len_seq, 1, 80, 15)
# initialize the model
model_keras_cnn_0 = jan_original(filter_density=1,
dropout=0.5,
input_shape=input_shape,
batchNorm=False,
dense_activation='sigmoid',
channel=1,
stateful=stateful,
training=False,
bidi=varin['bidi'])
model_keras_cnn_0.load_weights(full_path_model + str(ii) + '.h5')
else:
model_keras_cnn_0 = None
if varin['dataset'] != 'ismir':
# evaluate nacta 2017 dataset
batch_process_onset_detection(wav_path=nacta2017_wav_path,
textgrid_path=nacta2017_textgrid_path,
score_path=nacta2017_score_pinyin_path,
test_recordings=test_nacta_2017,
model_keras_cnn_0=model_keras_cnn_0,
cnnModel_name=model_name + str(ii),
detection_results_path=detection_results_path + str(ii),
scaler=scaler,
threshold=th,
obs_cal=obs_cal,
decoding_method='peakPicking',
architecture=architecture,
len_seq=len_seq,
stateful=stateful)
# evaluate nacta dataset
eval_results_decoding_path = \
batch_process_onset_detection(wav_path=nacta_wav_path,
textgrid_path=nacta_textgrid_path,
score_path=nacta_score_pinyin_path,
test_recordings=test_nacta,
model_keras_cnn_0=model_keras_cnn_0,
cnnModel_name=model_name + str(ii),
detection_results_path=detection_results_path + str(ii),
scaler=scaler,
threshold=th,
obs_cal=obs_cal,
decoding_method='peakPicking',
stateful=stateful,
architecture=architecture,
len_seq=len_seq)
append_write = append_or_write(eval_result_file_name)
with open(eval_result_file_name, append_write) as testfile:
csv_writer = csv.writer(testfile)
csv_writer.writerow([th])
precision_onset, recall_onset, F1_onset, \
precision, recall, F1, \
= eval_write_2_txt(eval_result_file_name,
eval_results_decoding_path,
label=False,
decoding_method='peakPicking')
list_precision_onset_25.append(precision_onset[0])
list_precision_onset_5.append(precision_onset[1])
list_recall_onset_25.append(recall_onset[0])
list_recall_onset_5.append(recall_onset[1])
list_F1_onset_25.append(F1_onset[0])
list_F1_onset_5.append(F1_onset[1])
list_precision_25.append(precision[0])
list_precision_5.append(precision[1])
list_recall_25.append(recall[0])
list_recall_5.append(recall[1])
list_F1_25.append(F1[0])
list_F1_5.append(F1[1])
return list_precision_onset_25, \
list_recall_onset_25, \
list_F1_onset_25, \
list_precision_25, \
list_recall_25, \
list_F1_25, \
list_precision_onset_5, \
list_recall_onset_5, \
list_F1_onset_5, \
list_precision_5, \
list_recall_5, \
list_F1_5
def viterbi_label_eval(test_nacta_2017,
test_nacta,
eval_label,
obs_cal,
len_seq,
model_name,
scaler,
full_path_model,
architecture,
detection_results_path,
jingju_eval_results_path):
"""evaluate viterbi decoding results"""
list_precision_onset_25, list_recall_onset_25, list_F1_onset_25, list_precision_25, list_recall_25, list_F1_25, \
list_precision_onset_5, list_recall_onset_5, list_F1_onset_5, list_precision_5, list_recall_5, list_F1_5 = \
viterbi_subroutine(test_nacta_2017=test_nacta_2017,
test_nacta=test_nacta,
eval_label=eval_label,
obs_cal=obs_cal,
len_seq=len_seq,
model_name=model_name,
architecture=architecture,
scaler=scaler,
full_path_model=full_path_model,
detection_results_path=detection_results_path)
postfix_statistic_sig = 'label' if eval_label else 'nolabel'
pickle.dump(list_F1_onset_25,
open(join('./statisticalSignificance/data/jingju',
varin['sample_weighting'],
model_name + '_' + 'viterbi' + '_' + postfix_statistic_sig + '.pkl'), 'w'))
write_results_2_txt_jingju(join(jingju_eval_results_path,
varin['sample_weighting'],
model_name + '_viterbi' + '_' + postfix_statistic_sig + '.txt'),
postfix_statistic_sig,
'viterbi',
list_precision_onset_25,
list_recall_onset_25,
list_F1_onset_25,
list_precision_25,
list_recall_25,
list_F1_25,
list_precision_onset_5,
list_recall_onset_5,
list_F1_onset_5,
list_precision_5,
list_recall_5,
list_F1_5)
def peak_picking_eval(test_nacta_2017,
test_nacta,
obs_cal,
len_seq,
model_name,
scaler,
full_path_model,
architecture,
detection_results_path,
jingju_eval_results_path):
"""evaluate the peak picking results"""
# coarse search
best_F1_onset_25, best_th = 0, 0
for th in range(1, 9):
th *= 0.1
try:
_, _, list_F1_onset_25, _, _, _, _, _, _, _, _, _ = \
peak_picking_subroutine(test_nacta_2017=test_nacta_2017,
test_nacta=test_nacta,
th=th,
obs_cal=obs_cal,
scaler=scaler,
len_seq=len_seq,
model_name=model_name,
full_path_model=full_path_model,
architecture=architecture,
detection_results_path=detection_results_path,
jingju_eval_results_path=jingju_eval_results_path)
if np.mean(list_F1_onset_25) > best_F1_onset_25:
best_th = th
best_F1_onset_25 = np.mean(list_F1_onset_25)
except:
continue
# finer scan the best threshold
for th in range(int((best_th - 0.1) * 100), int((best_th + 0.1) * 100)):
th *= 0.01
_, _, list_F1_onset_25, _, _, _, _, _, _, _, _, _ = \
peak_picking_subroutine(test_nacta_2017=test_nacta_2017,
test_nacta=test_nacta,
th=th,
obs_cal=obs_cal,
scaler=scaler,
len_seq=len_seq,
model_name=model_name,
full_path_model=full_path_model,
architecture=architecture,
detection_results_path=detection_results_path,
jingju_eval_results_path=jingju_eval_results_path)
if np.mean(list_F1_onset_25) > best_F1_onset_25:
best_th = th
best_F1_onset_25 = np.mean(list_F1_onset_25)
# get the statistics of the best threshold
list_precision_onset_25, list_recall_onset_25, list_F1_onset_25, list_precision_25, list_recall_25, list_F1_25, \
list_precision_onset_5, list_recall_onset_5, list_F1_onset_5, list_precision_5, list_recall_5, list_F1_5 = \
peak_picking_subroutine(test_nacta_2017=test_nacta_2017,
test_nacta=test_nacta,
th=best_th,
obs_cal=obs_cal,
scaler=scaler,
len_seq=len_seq,
model_name=model_name,
full_path_model=full_path_model,
architecture=architecture,
detection_results_path=detection_results_path,
jingju_eval_results_path=jingju_eval_results_path)
print('best_th', best_th)
pickle.dump(list_F1_onset_25,
open(join('./statisticalSignificance/data/jingju',
varin['sample_weighting'],
model_name + '_peakPickingMadmom.pkl'), 'w'))
write_results_2_txt_jingju(
join(jingju_eval_results_path, varin['sample_weighting'],
model_name + '_peakPickingMadmom' + '.txt'),
str(best_th),
'peakPicking',
list_precision_onset_25,
list_recall_onset_25,
list_F1_onset_25,
list_precision_25,
list_recall_25,
list_F1_25,
list_precision_onset_5,
list_recall_onset_5,
list_F1_onset_5,
list_precision_5,
list_recall_5,
list_F1_5)
def run_process_jingju_crnn(architecture):
if architecture == 'bidi_lstms_100':
cnnModel_name = 'bidi_lstms_100'
len_seq = 100 # sub-sequence length
elif architecture == 'bidi_lstms_200':
cnnModel_name = 'bidi_lstms_200'
len_seq = 200
elif architecture == 'bidi_lstms_400':
cnnModel_name = 'bidi_lstms_400'
len_seq = 400
else:
raise ValueError('There is no such architecture %s for CRNN.' % architecture)
scaler_artist_filter_phrase_model_path = join(jingju_cnn_model_path, 'scaler_jingju_crnn_phrase.pkl')
detection_results_path = join(root_path, 'eval', 'results', cnnModel_name)
jingju_eval_results_path = join(root_path, 'eval', 'jingju', 'results')
full_path_model = join(jingju_cnn_model_path, cnnModel_name)
test_nacta_2017, test_nacta = getTestRecordingsScoreDurCorrectionArtistAlbumFilter()
scaler = pickle.load(open(scaler_artist_filter_phrase_model_path, 'rb'))
obs_cal = 'tocal'
viterbi_label_eval(test_nacta_2017=test_nacta_2017,
test_nacta=test_nacta,
eval_label=True,
obs_cal=obs_cal,
scaler=scaler,
len_seq=len_seq,
model_name=cnnModel_name,
full_path_model=full_path_model,
architecture=architecture,
detection_results_path=detection_results_path,
jingju_eval_results_path=jingju_eval_results_path)
obs_cal = 'toload'
viterbi_label_eval(test_nacta_2017=test_nacta_2017,
test_nacta=test_nacta,
eval_label=False,
obs_cal=obs_cal,
scaler=scaler,
len_seq=len_seq,
model_name=cnnModel_name,
full_path_model=full_path_model,
architecture=architecture,
detection_results_path=detection_results_path,
jingju_eval_results_path=jingju_eval_results_path)
# peak picking evaluation
peak_picking_eval(test_nacta_2017=test_nacta_2017,
test_nacta=test_nacta,
obs_cal=obs_cal,
scaler=scaler,
len_seq=len_seq,
model_name=cnnModel_name,
full_path_model=full_path_model,
architecture=architecture,
detection_results_path=detection_results_path,
jingju_eval_results_path=jingju_eval_results_path)