-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathbenchmark_human_recordings.py
515 lines (418 loc) · 19.9 KB
/
benchmark_human_recordings.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
"""
The human recordings benchmark compares the resynthesis quality for a small set
of words (Problem, Studium, Wissenschaft, Liebe, genau, Klasse for now) between
the recovering of the segment based synthesis and the resynthesis of the human
recording.
"""
import pickle
import os
import time
import pandas as pd
from tqdm import tqdm
from paule import grad_plan
from paule import util
from paule import models
import torch
import numpy as np
tqdm.pandas()
torch.set_num_threads(4)
CONDITION = 'default'
#CONDITION = 'init-seg_baseline'
#CONDITION = 'pred_baseline'
#CONDITION = 'baseline_segment-embedder'
#CONDITION = 'segment-embedder'
if CONDITION == 'default':
paule = grad_plan.Paule(device=torch.device('cpu'))
elif CONDITION == 'pred_baseline':
pred_model = models.Non_Linear_Model(mode='pred', on_full_sequence=True).double()
pred_model.load_state_dict(torch.load("../paule/paule/pretrained_models/baseline_nonlinear/non_linear_pred_model_fullseq_8192_lr_0001_50_00001_50_000001_50_0000001_200.pt", map_location=torch.device('cpu')))
paule = grad_plan.Paule(pred_model=pred_model, device=torch.device('cpu'))
elif CONDITION == 'init-seg_baseline':
pred_model = models.Non_Linear_Model(mode='pred', on_full_sequence=True).double()
pred_model.load_state_dict(torch.load("../paule/paule/pretrained_models/baseline_nonlinear/non_linear_pred_model_fullseq_8192_lr_0001_50_00001_50_000001_50_0000001_200.pt", map_location=torch.device('cpu')))
paule = grad_plan.Paule(pred_model=pred_model, device=torch.device('cpu'))
elif CONDITION == 'baseline_segment-embedder':
pred_model = models.Non_Linear_Model(mode='pred', on_full_sequence=True).double()
pred_model.load_state_dict(torch.load("../paule/paule/pretrained_models/baseline_nonlinear/non_linear_pred_model_fullseq_8192_lr_0001_50_00001_50_000001_50_0000001_200.pt", map_location=torch.device('cpu')))
embedder = models.MelEmbeddingModel_MelSmoothResidualUpsampling(mel_smooth_layers=3).double()
embedder.load_state_dict(torch.load("../paule/paule/pretrained_models/embedder/model_synthesized_embed_model_3_4_180_8192_rmse_lr_00001_400.pt", map_location=torch.device('cpu')))
paule = grad_plan.Paule(pred_model=pred_model, embedder=embedder, device=torch.device('cpu'))
elif CONDITION == 'segment-embedder':
embedder = models.MelEmbeddingModel_MelSmoothResidualUpsampling(mel_smooth_layers=3).double()
embedder.load_state_dict(torch.load("../paule/paule/pretrained_models/embedder/model_synthesized_embed_model_3_4_180_8192_rmse_lr_00001_400.pt", map_location=torch.device('cpu')))
paule = grad_plan.Paule(embedder=embedder, device=torch.device('cpu'))
else:
raise ValueError('condition is wrong')
RESULT_DIR = f'results_{time.strftime("%Y%m%d")}_{CONDITION}/'
os.mkdir(RESULT_DIR)
dat = pd.read_pickle('data/geco_df_test_final_subset.pkl')
N_INNER = 100
N_OUTER = 40
dat = dat[-35:]
## for testing run it on the first two
#dat = dat[3:5]
dat = dat.iloc[[90, 100, 120]]
#N_INNER = 20
#N_OUTER = 2
if 'total_count_train_valid' in dat.columns:
dat_raw = dat
dat = dat_raw[['label', 'file_rec',
'vector',
'wav_rec', 'sampling_rate_rec', 'melspec_norm_rec', 'melspec_norm_min_rec',
'cp_norm',
'wav_syn', 'sampling_rate_syn', 'melspec_norm_syn', 'melspec_norm_min_syn']]
dat.columns = ['label', 'file', 'vector', 'rec_sig', 'rec_sr',
'rec_mel', 'rec_mel_min', 'seg_cp', 'seg_sig', 'seg_sr', 'seg_mel',
'seg_mel_min']
else:
# add segment synthesis
dat['seg_sig'] = dat['seg_cp'].progress_apply(lambda cp: util.speak(util.inv_normalize_cp(cp))[0])
dat['seg_sr'] = 44100
dat['seg_mel'] = dat['seg_sig'].progress_apply(lambda sig: util.normalize_mel_librosa(util.librosa_melspec(sig, 44100))) # WARNING this only works for synthesised audio as the sample rate is always 44100
dat['seg_mel_min'] = dat['seg_mel'].apply(lambda mel: mel.min())
dat['seg_mel'] = dat['seg_mel'] - dat['seg_mel_min']
# 1. Acoustic resynthesis
# 1.1 only acoustic objective
acoustic_results = list()
planned_cps = list()
inv_cps = list()
prod_signals = list()
prod_mels = list()
for index, row in tqdm(dat.iterrows(), total=dat.shape[0], desc='acoustic_acoustics'):
target_sig = row['rec_sig'].copy()
target_sig /= np.max(np.abs(target_sig))
target_sig *= 0.1
target_semvec = torch.tensor(row['vector'].copy()).view(1, 300)
result = paule.plan_resynth(
target_acoustic=(target_sig, row['rec_sr']),
target_semvec=target_semvec,
initialize_from='acoustic',
objective="acoustic",
log_semantics=True, n_inner=N_INNER, n_outer=N_OUTER)
acoustic_results.append(result)
(planned_cp, inv_cp, target_sig, target_mel, prod_sig, prod_mel, pred_mel, loss_steps,
loss_mel_steps, loss_semvec_steps, loss_jerk_steps, loss_velocity_steps, loss_prod_steps) = result
planned_cps.append(planned_cp)
inv_cps.append(inv_cp)
prod_signals.append(prod_sig)
prod_mels.append(prod_mel)
print(f"last semvec loss: {loss_semvec_steps[-1]:.2e}; last mel loss: {loss_mel_steps[-1]:.2e}\n\n")
dat['inv_cp'] = None
dat['inv_cp'] = inv_cps
dat['planned_cp_acoustic'] = None
dat['prod_sig_acoustic'] = None
dat['prod_mel_acoustic'] = None
dat['planned_cp_acoustic'] = planned_cps
dat['prod_sig_acoustic'] = prod_signals
dat['prod_mel_acoustic'] = prod_mels
# add inv synthesis
dat['inv_sig'] = dat['inv_cp'].progress_apply(lambda cp: util.speak(util.inv_normalize_cp(cp))[0])
dat['inv_sr'] = 44100
dat['inv_mel'] = dat['inv_sig'].progress_apply(lambda sig: util.normalize_mel_librosa(util.librosa_melspec(sig, 44100))) # WARNING this only works for synthesised audio as the sample rate is always 44100
dat['inv_mel_min'] = dat['inv_mel'].apply(lambda mel: mel.min())
dat['inv_mel'] = dat['inv_mel'] - dat['inv_mel_min']
with open(os.path.join(RESULT_DIR, 'acoustics_acoustic_results.pickle'), 'wb') as pfile:
pickle.dump(acoustic_results, pfile)
# 1.2 semantic and acoustic objective
acoustic_semvec_results = list()
planned_cps = list()
prod_signals = list()
prod_mels = list()
inv_cps = list()
for index, row in tqdm(dat.iterrows(), total=dat.shape[0], desc='acoustic_acoustic-semvec'):
target_sig = row['rec_sig'].copy()
target_sig /= np.max(np.abs(target_sig))
target_sig *= 0.1
target_semvec = torch.tensor(row['vector'].copy()).view(1, 300)
result = paule.plan_resynth(
target_acoustic=(target_sig, row['rec_sr']),
target_semvec=target_semvec,
initialize_from='acoustic',
objective="acoustic_semvec",
log_semantics=True, n_inner=N_INNER, n_outer=N_OUTER)
acoustic_semvec_results.append(result)
(planned_cp, inv_cp, target_sig, target_mel, prod_sig, prod_mel, pred_mel, loss_steps,
loss_mel_steps, loss_semvec_steps, loss_jerk_steps, loss_velocity_steps, loss_prod_steps) = result
planned_cps.append(planned_cp)
prod_signals.append(prod_sig)
prod_mels.append(prod_mel)
inv_cps.append(inv_cp)
print(f"last semvec loss: {loss_semvec_steps[-1]:.2e}; last mel loss: {loss_mel_steps[-1]:.2e}")
dat['planned_cp_acoustic-semvec'] = None
dat['prod_sig_acoustic-semvec'] = None
dat['prod_mel_acoustic-semvec'] = None
dat['planned_cp_acoustic-semvec'] = planned_cps
dat['prod_sig_acoustic-semvec'] = prod_signals
dat['prod_mel_acoustic-semvec'] = prod_mels
with open(os.path.join(RESULT_DIR, 'acoustics_acoustic-semvec_results.pickle'), 'wb') as pfile:
pickle.dump(acoustic_semvec_results, pfile)
# 1.3 only semantic objective
semvec_results = list()
planned_cps = list()
prod_signals = list()
prod_mels = list()
inv_cps = list()
for index, row in tqdm(dat.iterrows(), total=dat.shape[0], desc='acoustic_semvec'):
target_sig = row['rec_sig'].copy()
target_sig /= np.max(np.abs(target_sig))
target_sig *= 0.1
target_semvec = torch.tensor(row['vector'].copy()).view(1, 300)
result = paule.plan_resynth(
target_acoustic=(target_sig, row['rec_sr']),
target_semvec=target_semvec,
initialize_from='acoustic',
objective="semvec",
log_semantics=True, n_inner=N_INNER, n_outer=N_OUTER)
semvec_results.append(result)
(planned_cp, inv_cp, target_sig, target_mel, prod_sig, prod_mel, pred_mel, loss_steps,
loss_mel_steps, loss_semvec_steps, loss_jerk_steps, loss_velocity_steps, loss_prod_steps) = result
planned_cps.append(planned_cp)
prod_signals.append(prod_sig)
prod_mels.append(prod_mel)
inv_cps.append(inv_cp)
print(f"last semvec loss: {loss_semvec_steps[-1]:.2e}; last mel loss: {loss_mel_steps[-1]:.2e}")
dat['planned_cp_semvec'] = None
dat['prod_sig_semvec'] = None
dat['prod_mel_semvec'] = None
dat['planned_cp_semvec'] = planned_cps
dat['prod_sig_semvec'] = prod_signals
dat['prod_mel_semvec'] = prod_mels
with open(os.path.join(RESULT_DIR, 'acoustics_semvec_results.pickle'), 'wb') as pfile:
pickle.dump(semvec_results, pfile)
dat.to_pickle(os.path.join(RESULT_DIR, 'dat_acoustics.pickle'))
# TODO save results and data
# 1.4 Conditions
# segment based synthesis
# inverse prduction
# (after first planning cycle)
# after 40 planning cycles
# 1.5 Metrics
# rmse log-mel-spectrogramm from production compared to target acoustics
# Wasserstein distance log-mel-spectrogramm from production compared to target acoustics
# rmse semantic vector from production to target acoustics
# classification rank
# 2. Semvec resyntesis
# 2.1 acoustic semvec objective
vector_results = list()
planned_cps = list()
inv_cps = list()
prod_signals = list()
prod_mels = list()
for index, row in tqdm(dat.iterrows(), total=dat.shape[0], desc='vector_acoustics-semvec'):
target_seq_length = row['rec_mel'].shape[0]
target_semvec = torch.tensor(row['vector'].copy()).view(1, 300)
result = paule.plan_resynth(
target_semvec=target_semvec,
target_seq_length=target_seq_length,
initialize_from='semvec',
objective="acoustic_semvec",
log_semantics=True, n_inner=N_INNER, n_outer=N_OUTER, seed=20210811 + index)
vector_results.append(result)
(planned_cp, inv_cp, target_sig, target_mel, prod_sig, prod_mel, pred_mel, loss_steps,
loss_mel_steps, loss_semvec_steps, loss_jerk_steps, loss_velocity_steps, loss_prod_steps) = result
planned_cps.append(planned_cp)
inv_cps.append(inv_cp)
prod_signals.append(prod_sig)
prod_mels.append(prod_mel)
print(f"last semvec loss: {loss_semvec_steps[-1]:.2e}; last mel loss: {loss_mel_steps[-1]:.2e}\n\n")
dat['planned_cp_vector_acoustic-semvec'] = None
dat['prod_sig_vector_acoustic-semvec'] = None
dat['prod_mel_vector_acoustic-semvec'] = None
dat['planned_cp_vector_acoustic-semvec'] = planned_cps
dat['prod_sig_vector_acoustic-semvec'] = prod_signals
dat['prod_mel_vector_acoustic-semvec'] = prod_mels
with open(os.path.join(RESULT_DIR, 'vector_acoustic-semvec_results.pickle'), 'wb') as pfile:
pickle.dump(vector_results, pfile)
dat['gan_cp'] = None
dat['gan_cp'] = inv_cps
# add gan synthesis
dat['gan_sig'] = dat['gan_cp'].progress_apply(lambda cp: util.speak(util.inv_normalize_cp(cp))[0])
dat['gan_sr'] = 44100
dat['gan_mel'] = dat['gan_sig'].progress_apply(lambda sig: util.normalize_mel_librosa(util.librosa_melspec(sig, 44100))) # NOTE: this only works for synthesised audio as the sample rate is always 44100
dat['gan_mel_min'] = dat['gan_mel'].apply(lambda mel: mel.min())
dat['gan_mel'] = dat['gan_mel'] - dat['gan_mel_min']
# 2.2 acoustic only objective
vector_results = list()
planned_cps = list()
inv_cps = list()
prod_signals = list()
prod_mels = list()
for index, row in tqdm(dat.iterrows(), total=dat.shape[0], desc='vector_acoustics'):
target_seq_length = row['rec_mel'].shape[0]
target_semvec = torch.tensor(row['vector'].copy()).view(1, 300)
result = paule.plan_resynth(
target_semvec=target_semvec,
target_seq_length=target_seq_length,
initialize_from='semvec',
objective="acoustic",
log_semantics=True, n_inner=N_INNER, n_outer=N_OUTER, seed=20210811 + index)
vector_results.append(result)
(planned_cp, inv_cp, target_sig, target_mel, prod_sig, prod_mel, pred_mel, loss_steps,
loss_mel_steps, loss_semvec_steps, loss_jerk_steps, loss_velocity_steps, loss_prod_steps) = result
planned_cps.append(planned_cp)
inv_cps.append(inv_cp)
prod_signals.append(prod_sig)
prod_mels.append(prod_mel)
print(f"last semvec loss: {loss_semvec_steps[-1]:.2e}; last mel loss: {loss_mel_steps[-1]:.2e}\n\n")
dat['planned_cp_vector_acoustic'] = None
dat['prod_sig_vector_acoustic'] = None
dat['prod_mel_vector_acoustic'] = None
dat['planned_cp_vector_acoustic'] = planned_cps
dat['prod_sig_vector_acoustic'] = prod_signals
dat['prod_mel_vector_acoustic'] = prod_mels
with open(os.path.join(RESULT_DIR, 'vector_acoustic_results.pickle'), 'wb') as pfile:
pickle.dump(vector_results, pfile)
# 2.3 semvec only objective
vector_results = list()
planned_cps = list()
inv_cps = list()
prod_signals = list()
prod_mels = list()
for index, row in tqdm(dat.iterrows(), total=dat.shape[0], desc='vector_semvec'):
target_seq_length = row['rec_mel'].shape[0]
target_semvec = torch.tensor(row['vector'].copy()).view(1, 300)
result = paule.plan_resynth(
target_semvec=target_semvec,
target_seq_length=target_seq_length,
initialize_from='semvec',
objective="semvec",
log_semantics=True, n_inner=N_INNER, n_outer=N_OUTER, seed=20210811 + index)
vector_results.append(result)
(planned_cp, inv_cp, target_sig, target_mel, prod_sig, prod_mel, pred_mel, loss_steps,
loss_mel_steps, loss_semvec_steps, loss_jerk_steps, loss_velocity_steps, loss_prod_steps) = result
planned_cps.append(planned_cp)
inv_cps.append(inv_cp)
prod_signals.append(prod_sig)
prod_mels.append(prod_mel)
print(f"last semvec loss: {loss_semvec_steps[-1]:.2e}; last mel loss: {loss_mel_steps[-1]:.2e}\n\n")
dat['planned_cp_vector_semvec'] = None
dat['prod_sig_vector_semvec'] = None
dat['prod_mel_vector_semvec'] = None
dat['planned_cp_vector_semvec'] = planned_cps
dat['prod_sig_vector_semvec'] = prod_signals
dat['prod_mel_vector_semvec'] = prod_mels
with open(os.path.join(RESULT_DIR, 'vector_semvec.pickle'), 'wb') as pfile:
pickle.dump(vector_results, pfile)
dat.to_pickle(os.path.join(RESULT_DIR, 'dat_vector.pickle'))
# 3 Initialise with segment based cps
# 3.1 only acoustic objective
acoustic_results = list()
planned_cps = list()
inv_cps = list()
prod_signals = list()
prod_mels = list()
for index, row in tqdm(dat.iterrows(), total=dat.shape[0], desc='init-seg_acoustic'):
target_sig = row['rec_sig'].copy()
target_sig /= np.max(np.abs(target_sig))
target_sig *= 0.1
target_semvec = torch.tensor(row['vector'].copy()).view(1, 300)
inv_cp = row['seg_cp'].copy()
seg_padding = int((row['seg_cp'].shape[0] - row['rec_mel'].shape[0] * 2) / 2)
inv_cp = inv_cp[seg_padding:-seg_padding]
result = paule.plan_resynth(
target_acoustic=(target_sig, row['rec_sr']),
target_semvec=target_semvec,
inv_cp=inv_cp,
initialize_from=None,
objective="acoustic",
log_semantics=True, n_inner=N_INNER, n_outer=N_OUTER)
acoustic_results.append(result)
(planned_cp, inv_cp, target_sig, target_mel, prod_sig, prod_mel, pred_mel, loss_steps,
loss_mel_steps, loss_semvec_steps, loss_jerk_steps, loss_velocity_steps, loss_prod_steps) = result
planned_cps.append(planned_cp)
inv_cps.append(inv_cp)
prod_signals.append(prod_sig)
prod_mels.append(prod_mel)
print(f"last semvec loss: {loss_semvec_steps[-1]:.2e}; last mel loss: {loss_mel_steps[-1]:.2e}\n\n")
dat['inv_cp'] = None
dat['inv_cp'] = inv_cps
dat['planned_cp_acoustic'] = None
dat['prod_sig_acoustic'] = None
dat['prod_mel_acoustic'] = None
dat['planned_cp_acoustic'] = planned_cps
dat['prod_sig_acoustic'] = prod_signals
dat['prod_mel_acoustic'] = prod_mels
# add inv synthesis
dat['inv_sig'] = dat['inv_cp'].progress_apply(lambda cp: util.speak(util.inv_normalize_cp(cp))[0])
dat['inv_sr'] = 44100
dat['inv_mel'] = dat['inv_sig'].progress_apply(lambda sig: util.normalize_mel_librosa(util.librosa_melspec(sig, 44100))) # WARNING this only works for synthesised audio as the sample rate is always 44100
dat['inv_mel_min'] = dat['inv_mel'].apply(lambda mel: mel.min())
dat['inv_mel'] = dat['inv_mel'] - dat['inv_mel_min']
with open(os.path.join(RESULT_DIR, 'init-seg_acoustic_results.pickle'), 'wb') as pfile:
pickle.dump(acoustic_results, pfile)
# 3.2 semantic and acoustic objective
acoustic_semvec_results = list()
planned_cps = list()
prod_signals = list()
prod_mels = list()
inv_cps = list()
for index, row in tqdm(dat.iterrows(), total=dat.shape[0], desc='init-seg_acoustic-semvec'):
target_sig = row['rec_sig'].copy()
target_sig /= np.max(np.abs(target_sig))
target_sig *= 0.1
target_semvec = torch.tensor(row['vector'].copy()).view(1, 300)
inv_cp = row['seg_cp'].copy()
seg_padding = int((row['seg_cp'].shape[0] - row['rec_mel'].shape[0] * 2) / 2)
inv_cp = inv_cp[seg_padding:-seg_padding]
result = paule.plan_resynth(
target_acoustic=(target_sig, row['rec_sr']),
target_semvec=target_semvec,
inv_cp=inv_cp,
initialize_from=None,
objective="acoustic_semvec",
log_semantics=True, n_inner=N_INNER, n_outer=N_OUTER)
acoustic_semvec_results.append(result)
(planned_cp, inv_cp, target_sig, target_mel, prod_sig, prod_mel, pred_mel, loss_steps,
loss_mel_steps, loss_semvec_steps, loss_jerk_steps, loss_velocity_steps, loss_prod_steps) = result
planned_cps.append(planned_cp)
prod_signals.append(prod_sig)
prod_mels.append(prod_mel)
inv_cps.append(inv_cp)
print(f"last semvec loss: {loss_semvec_steps[-1]:.2e}; last mel loss: {loss_mel_steps[-1]:.2e}")
dat['planned_cp_acoustic-semvec'] = None
dat['prod_sig_acoustic-semvec'] = None
dat['prod_mel_acoustic-semvec'] = None
dat['planned_cp_acoustic-semvec'] = planned_cps
dat['prod_sig_acoustic-semvec'] = prod_signals
dat['prod_mel_acoustic-semvec'] = prod_mels
with open(os.path.join(RESULT_DIR, 'init-seg_acoustic-semvec_results.pickle'), 'wb') as pfile:
pickle.dump(acoustic_semvec_results, pfile)
# 3.3 only semantic objective
semvec_results = list()
planned_cps = list()
prod_signals = list()
prod_mels = list()
inv_cps = list()
for index, row in tqdm(dat.iterrows(), total=dat.shape[0], desc='init-seg_semvec'):
target_sig = row['rec_sig'].copy()
target_sig /= np.max(np.abs(target_sig))
target_sig *= 0.1
target_semvec = torch.tensor(row['vector'].copy()).view(1, 300)
inv_cp = row['seg_cp'].copy()
seg_padding = int((row['seg_cp'].shape[0] - row['rec_mel'].shape[0] * 2) / 2)
inv_cp = inv_cp[seg_padding:-seg_padding]
result = paule.plan_resynth(
target_acoustic=(target_sig, row['rec_sr']),
target_semvec=target_semvec,
inv_cp=inv_cp,
initialize_from=None,
objective="semvec",
log_semantics=True, n_inner=N_INNER, n_outer=N_OUTER)
semvec_results.append(result)
(planned_cp, inv_cp, target_sig, target_mel, prod_sig, prod_mel, pred_mel, loss_steps,
loss_mel_steps, loss_semvec_steps, loss_jerk_steps, loss_velocity_steps, loss_prod_steps) = result
planned_cps.append(planned_cp)
prod_signals.append(prod_sig)
prod_mels.append(prod_mel)
inv_cps.append(inv_cp)
print(f"last semvec loss: {loss_semvec_steps[-1]:.2e}; last mel loss: {loss_mel_steps[-1]:.2e}")
dat['planned_cp_semvec'] = None
dat['prod_sig_semvec'] = None
dat['prod_mel_semvec'] = None
dat['planned_cp_semvec'] = planned_cps
dat['prod_sig_semvec'] = prod_signals
dat['prod_mel_semvec'] = prod_mels
with open(os.path.join(RESULT_DIR, 'init-seg_semvec_results.pickle'), 'wb') as pfile:
pickle.dump(semvec_results, pfile)
dat.to_pickle(os.path.join(RESULT_DIR, 'dat_init-seg.pickle'))
# 4. "Wer ist eigentlich Paule?"