Mercurial > hg > chourdakisreiss2018smc

annotate demo/workspace/approach1.py @ 13:16066f0a7127 tip

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
fixed the problem with brat
author Emmanouil Theofanis Chourdakis <e.t.chourdakis@qmul.ac.uk>
date Sat, 08 Dec 2018 11:02:40 +0000
parents 90155bdd5dd6
children
rev   line source
e@0 1 import difflib
e@0 2 from sklearn.externals import joblib
e@0 3 from collections import defaultdict
e@0 4 import nltk
e@0 5 import numpy as np
e@0 6 import re
e@0 7 import librosa
e@0 8 import glob
e@0 9 import pandas as pd
e@0 10 from nltk.stem import porter
e@0 11 import sox
e@0 12 from scipy.io.wavfile import read as wavread
e@0 13 from scipy.io.wavfile import write as wavwrite
e@0 14 from numpy.core._internal import _gcd as gcd
e@0 15 from rtsfx import *
e@0 16
e@0 17 import subprocess
e@0 18 import os
e@0 19 import pypeg2 as pg
e@0 20 import random
e@0 21
e@0 22 VOICE_PROPERTIES = ['slow', 'deep', 'fast', 'stuttering']
e@0 23 PAUSE_PROPERTIES = ['short', 'long']
e@0 24 SEPARATORS = [ "[", "]", "(", ")", ":", "-"]
e@0 25 SFX_MOD_PROPERTIES = ['quiet', 'loud', 'silent']
e@0 26
e@0 27
e@0 28 FIXED_VOICES = False
e@0 29
e@0 30 FMV = 0
e@0 31 FFV = 0
e@0 32
e@0 33
e@0 34 male_voices = r"""
e@0 35 cmu_us_ahw_cg
e@0 36 cmu_us_awb_cg
e@0 37 cmu_us_bdl_cg
e@0 38 cmu_us_fem_cg
e@0 39 cmu_us_jmk_cg
e@0 40 cmu_us_ksp_cg
e@0 41 cmu_us_rms_cg
e@0 42 """.split()
e@0 43
e@0 44
e@0 45 female_voices = r"""
e@0 46 cmu_us_aup_cg
e@0 47 cmu_us_axb_cg
e@0 48 cmu_us_clb_cg
e@0 49 cmu_us_gka_cg
e@0 50 cmu_us_rxr_cg
e@0 51 cmu_us_slt_cg
e@0 52 """.split()
e@0 53
e@0 54
e@0 55
e@0 56 # male_voices = r"""
e@0 57 # cmu_us_ahw_cg
e@0 58 # cmu_us_fem_cg
e@0 59 # cmu_us_rms_cg
e@0 60 # """.split()
e@0 61 #
e@0 62 #
e@0 63 # female_voices = r"""
e@0 64 # cmu_us_aup_cg
e@0 65 # cmu_us_axb_cg
e@0 66 # cmu_us_rxr_cg
e@0 67 # cmu_us_slt_cg
e@0 68 # """.split()
e@0 69
e@0 70 fixed_male_voice = male_voices
e@0 71 fixed_female_voice = female_voices
e@0 72
e@0 73
e@0 74 # male_voices = r"""
e@0 75 # cmu_us_ahw_cg
e@0 76 # cmu_us_awb_cg
e@0 77 # cmu_us_bdl_cg
e@0 78 # cmu_us_fem_cg
e@0 79 # cmu_us_jmk_cg
e@0 80 # cmu_us_ksp_cg
e@0 81 # cmu_us_rms_cg
e@0 82 # """.split()
e@0 83
e@0 84
e@0 85 # male_voices = r"""
e@0 86 # cmu_us_ahw_cg
e@0 87 # """.split()
e@0 88 #
e@0 89 #
e@0 90 # female_voices = r"""
e@0 91 # cmu_us_ahw_cg
e@0 92 # """.split()
e@0 93
e@0 94 import matplotlib.pyplot as plt
e@0 95
e@0 96 def generate_speech_with_festival(voice,
e@0 97 panning,
e@0 98 line,
e@0 99 sr=None
e@0 100 ):
e@0 101 """
e@0 102 Used for speech generation
e@0 103 Constructs a festival .sable file
e@0 104 and runs it through festival.
e@0 105
e@0 106 """
e@0 107 header = r"""
e@0 108 <?xml version="1.0"?>
e@0 109 <!DOCTYPE SABLE PUBLIC "-//SABLE//DTD SABLE speech mark up//EN"
e@0 110 "Sable.v0_2.dtd"
e@0 111 []>
e@0 112 <SABLE>
e@0 113 <SPEAKER NAME="{}">
e@0 114 """.format(voice)
e@0 115
e@0 116 footer = r"""
e@0 117 </SPEAKER>
e@0 118 </SABLE>
e@0 119 """
e@0 120
e@0 121 # 0. Construct sable file
e@0 122 sable = header + line + footer
e@0 123
e@0 124 # 1. Save sable file to a temporary .sable file in tmp
e@0 125
e@0 126 with open('/tmp/character_line.sable', 'w') as f:
e@0 127 f.write(sable)
e@0 128
e@0 129 # 2. Call process to festival
e@0 130 cmd = 'text2wave /tmp/character_line.sable -o /tmp/character_line.wav'
e@0 131
e@0 132 print("Generating speech for line: '{}' with voice '{}' and panning '{}' ".format(line, voice, panning))
e@0 133 value = subprocess.call(cmd, shell=True)
e@0 134
e@0 135 if value != 0:
e@0 136 raise RuntimeError("Festival failed to execute.")
e@0 137
e@0 138 # 3. Load back wave file
e@0 139 if sr is None:
e@0 140 wav, sr = librosa.load('/tmp/character_line.wav', mono=True)
e@0 141 else:
e@0 142 wav, sr = librosa.load('/tmp/character_line.wav', sr=sr, mono=True)
e@0 143
e@0 144 audio = np.vstack([panning*wav,(1.-panning)*wav])
e@0 145 #
e@0 146 # plt.figure()
e@0 147 # plt.plot(audio[0,:])
e@0 148 # plt.figure()
e@0 149 # plt.plot(audio[1,:])
e@0 150 # plt.show()
e@0 151 return audio, sr
e@0 152
e@0 153
e@0 154 def substr_features(sent,
e@0 155 lower=True, substr=[]):
e@0 156 if lower:
e@0 157 sent = sent.lower()
e@0 158 freqs = defaultdict(int)
e@0 159 for ss in substr:
e@0 160 if ss in sent:
e@0 161 freqs[ss] = 1
e@0 162 return dict(freqs)
e@0 163
e@0 164
e@0 165 def features_dict_to_matrix(features, feature_labels):
e@0 166 N = len(features)
e@0 167 M = len(feature_labels)
e@0 168 arr = np.zeros((N, M))
e@0 169
e@0 170 idx_to_feat = list(feature_labels)
e@0 171 feat_to_idx = dict((idx_to_feat[k], k) for k in range(len(idx_to_feat)))
e@0 172
e@0 173 for n in range(arr.shape[0]):
e@0 174 for m in range(arr.shape[1]):
e@0 175 if idx_to_feat[m] in features[n]:
e@0 176 arr[n, m] = features[n][idx_to_feat[m]]
e@0 177
e@0 178 return arr, list(feat_to_idx.keys())
e@0 179
e@0 180
e@0 181 def similar(text1, text2,
e@0 182 threshold=0.7 # threshold for similarity
e@0 183 ):
e@0 184 """ Tests whether two strings are similar """
e@0 185
e@0 186 ratio = difflib.SequenceMatcher(None, text1.lower(), text2.lower()).ratio()
e@0 187 return ratio >= threshold
e@0 188
e@0 189
e@0 190 class Master():
e@0 191 def __init__(self, downmix):
e@0 192 self.downmix = downmix
e@0 193
e@0 194
e@0 195 def get_mastered(self):
e@0 196 # Creating transformer
e@0 197 tfm = sox.Transformer()
e@0 198
e@0 199 # Removing everything below 80hz
e@0 200 tfm.highpass(80)
e@0 201
e@0 202 # Adding a notch filter at 200hz to improve clarity
e@0 203 tfm.bandreject(200)
e@0 204
e@0 205 # Loudness control for under -9dB
e@0 206 tfm.loudness(gain_db=-9)
e@0 207
e@0 208 # Store downmix temporarily
e@0 209 librosa.output.write_wav('/tmp/downmix_unnormalized.wav', self.downmix, sr=44100, norm=False)
e@0 210 tfm.build('/tmp/downmix_unnormalized.wav', '/tmp/downmix_normalized.wav')
e@0 211
e@0 212 # Load downmix
e@0 213 mastered = librosa.core.load('/tmp/downmix_normalized.wav', sr=44100,mono=False )[0]
e@0 214 return mastered
e@0 215
e@0 216
e@0 217 class Mixer():
e@0 218 def __init__(self, multitrack):
e@0 219 self.multitrack = multitrack
e@0 220
e@0 221 def get_downmix(self):
e@0 222
e@0 223 # Just a trick to get the length of the first track
e@0 224 if 'background' in self.multitrack:
e@0 225 D = self.multitrack['background'].shape[1]
e@0 226 else:
e@0 227 for track in self.multitrack:
e@0 228 D = self.multitrack[track].shape[1]
e@0 229 break
e@0 230
e@0 231 downmix = np.zeros((2, D))
e@0 232 for ttrack in self.multitrack:
e@0 233
e@0 234 #1. Normalize
e@0 235
e@0 236 track = self.multitrack[ttrack]
e@0 237
e@0 238 max_val = np.max(np.abs(track))
e@0 239
e@0 240 if max_val > 0:
e@0 241 track /= max_val
e@0 242
e@0 243 if ttrack == 'background':
e@0 244 track *= 0.05
e@0 245
e@0 246
e@0 247 downmix += track
e@0 248
e@0 249 return downmix
e@0 250
e@0 251
e@0 252 def zafar(lx, rx, d1, g1, m, fc, G, da=0.007, fs=44100.):
e@0 253 """ Rafii & Pardo Reverberator (2009) controlled by High Level parameters
e@0 254 Inputs:
e@0 255 lx : left channel input
e@0 256 rx : right channel input
e@0 257 d1 : delay of first comb filter in seconds
e@0 258 g1 : gain of first comb filters
e@0 259 da : delay of allpass filter in seconds
e@0 260 G : dry/wet mix gain
e@0 261 fc : lowpass filter cuttoff Hz
e@0 262 m : difference between left and right channel phases
e@0 263 fs : sampling rate
e@0 264
e@0 265 Outputs:
e@0 266 ly: left channel output
e@0 267 ry: right channel output
e@0 268 """
e@0 269
e@0 270 d1 = int(d1 * fs)
e@0 271 m = int(m * fs)
e@0 272 da = int(da * fs)
e@0 273
e@0 274 def calculate_parameters(d1, g1):
e@0 275
e@0 276 d2 = int(round((1.5) ** (-1) * d1))
e@0 277
e@0 278 while gcd(d2, d1) != 1:
e@0 279 d2 += 1
e@0 280
e@0 281 d3 = int(round((1.5) ** (-2) * d1))
e@0 282
e@0 283 while gcd(d3, d2) != 1 or gcd(d3, d1) != 1:
e@0 284 d3 += 1
e@0 285
e@0 286 d4 = int(round((1.5) ** (-3) * d1))
e@0 287
e@0 288 while gcd(d4, d3) != 1 or gcd(d4, d2) != 1 or gcd(d4, d1) != 1:
e@0 289 d4 += 1
e@0 290
e@0 291 d5 = int(round((1.5) ** (-4) * d1))
e@0 292
e@0 293 while gcd(d5, d4) != 1 or gcd(d5, d3) != 1 or gcd(d5, d2) != 1 or gcd(d5, d1) != 1:
e@0 294 d5 += 1
e@0 295
e@0 296 d6 = int(round((1.5) ** (-5) * d1))
e@0 297 while gcd(d6, d5) != 1 or gcd(d6, d4) != 1 or gcd(d6, d3) != 1 or gcd(d6, d2) != 1 or gcd(d6, d1) != 1:
e@0 298 d6 += 1
e@0 299 g2 = g1 ** (1.5) ** (-1) * g1
e@0 300 g3 = g1 ** (1.5) ** (-2) * g1
e@0 301 g4 = g1 ** (1.5) ** (-3) * g1
e@0 302 g5 = g1 ** (1.5) ** (-4) * g1
e@0 303 g6 = g1 ** (1.5) ** (-5) * g1
e@0 304
e@0 305 return (d1, d2, d3, d4, d5, d6, g1, g2, g3, g4, g5, g6)
e@0 306
e@0 307 def comb_array(x, g1, d1):
e@0 308
e@0 309 (d1, d2, d3, d4, d5, d6, g1, g2, g3, g4, g5, g6) = calculate_parameters(d1, g1)
e@0 310
e@0 311 c1out = comb(x, g1, d1)
e@0 312 c2out = comb(x, g2, d2)
e@0 313 c3out = comb(x, g3, d3)
e@0 314 c4out = comb(x, g4, d4)
e@0 315 c5out = comb(x, g5, d5)
e@0 316 c6out = comb(x, g6, d6)
e@0 317
e@0 318 Lc1 = len(c1out)
e@0 319 Lc2 = len(c2out)
e@0 320 Lc3 = len(c3out)
e@0 321 Lc4 = len(c4out)
e@0 322 Lc5 = len(c5out)
e@0 323 Lc6 = len(c6out)
e@0 324
e@0 325 Lc = max(Lc1, Lc2, Lc3, Lc4, Lc5, Lc6)
e@0 326
e@0 327 y = np.zeros((Lc,))
e@0 328
e@0 329 y[0:Lc1] = c1out
e@0 330 y[0:Lc2] += c2out
e@0 331 y[0:Lc3] += c3out
e@0 332 y[0:Lc4] += c4out
e@0 333 y[0:Lc5] += c5out
e@0 334 y[0:Lc6] += c6out
e@0 335
e@0 336 return y
e@0 337
e@0 338 def comb(x, g, d):
e@0 339 LEN = len(x) + d
e@0 340 # print d
e@0 341 y = np.zeros((LEN,))
e@0 342 for n in range(0, LEN):
e@0 343 if n - d < 0:
e@0 344 y[n] = 0
e@0 345 else:
e@0 346 y[n] = x[n - d] + g * y[n - d]
e@0 347
e@0 348 return y
e@0 349
e@0 350 def allpass(x, g, d):
e@0 351 LENx = len(x)
e@0 352 LENy = LENx + d
e@0 353 y = np.zeros((LENy,))
e@0 354 for n in range(0, LENy):
e@0 355 if n - d < 0:
e@0 356 y[n] = -g * x[n]
e@0 357 elif n >= LENx:
e@0 358 y[n] = x[n - d] + g * y[n - d]
e@0 359 else:
e@0 360 y[n] = x[n - d] - g * x[n] + g * y[n - d]
e@0 361
e@0 362 return y
e@0 363
e@0 364 def lowpass(x, g):
e@0 365 LEN = len(x)
e@0 366 y = np.zeros((LEN,))
e@0 367
e@0 368 for n in range(0, LEN):
e@0 369 if n - 1 < 0:
e@0 370 y[n] = (1 - g) * x[n]
e@0 371 else:
e@0 372 y[n] = (1 - g) * x[n] + g * y[n - 1]
e@0 373
e@0 374 return y
e@0 375
e@0 376 ga = 1. / np.sqrt(2.)
e@0 377
e@0 378 cin = 0.5 * lx + 0.5 * rx
e@0 379 cout = comb_array(cin, g1, d1)
e@0 380
e@0 381 ra = allpass(cout, ga, da + m // 2)
e@0 382 la = allpass(cout, ga, da - m // 2)
e@0 383
e@0 384 gc = 2 - np.cos(2 * np.pi * fc / fs) - np.sqrt((np.cos(2 * np.pi * fc / fs) - 2) ** 2 - 1)
e@0 385
e@0 386 ral = lowpass(ra, gc)
e@0 387 lal = lowpass(la, gc)
e@0 388
e@0 389 ralg = G * ral
e@0 390 lalg = G * lal
e@0 391
e@0 392 ry = ralg[0:len(rx)] + (1 - G) * rx
e@0 393 ly = lalg[0:len(lx)] + (1 - G) * lx
e@0 394
e@0 395 return np.vstack([ry, ly])
e@0 396
e@0 397 def get_reverb_from_tags(xl, xr, tags, fs=44100):
e@0 398 reverb_csv = 'contributions.csv'
e@0 399 df = pd.read_csv(reverb_csv)
e@0 400 df = df.fillna("")
e@0 401 params = []
e@0 402 for n in range(len(df)):
e@0 403 if all([t in df['agreed'].iloc[n].split(',') for t in tags]):
e@0 404 params.append(df['param'].iloc[n])
e@0 405 d1, g1, m, fc, G = [float(f) for f in params[0].split(',')]
e@0 406 y = zafar(xl, xr, d1, g1, m, fc, G, fs=fs)
e@0 407 return y
e@0 408
e@0 409
e@0 410 def fade(x, fade_in, fade_out, sr=44100):
e@0 411 """
e@0 412 Creates a fade-in-fade-out envelope
e@0 413 for audio array x.
e@0 414 """
e@0 415
e@0 416 if len(x) == 0:
e@0 417 return x
e@0 418
e@0 419 fade_in_samples = int(fade_in * sr)
e@0 420 fade_out_samples = int(fade_out * sr)
e@0 421
e@0 422 outp = np.ones_like(x)
e@0 423 for n in range(fade_in_samples):
e@0 424 outp[n] = n * 1. / fade_in_samples
e@0 425
e@0 426 for n in range(fade_out_samples):
e@0 427 outp[len(outp) - fade_out_samples + n] = 1 - 1. / fade_out_samples * n
e@0 428 return outp * x
e@0 429
e@0 430
e@0 431 def slope(x, slope_in, slope_out, delay=1.0, v=0.1, sr=44100):
e@0 432 """
e@0 433 Creates a slope in slope out envelope
e@0 434 """
e@0 435
e@0 436 if len(x) == 0:
e@0 437 return x
e@0 438
e@0 439 delay_samples = int(delay * sr)
e@0 440 slope_in_samples = int(slope_in * sr)
e@0 441 slope_out_samples = int(slope_out * sr)
e@0 442
e@0 443 outp = np.zeros_like(x)
e@0 444
e@0 445 for n in range(len(outp)):
e@0 446 if n >= 0 and n < delay_samples:
e@0 447 outp[n] = 1.0 - v
e@0 448 elif n >= delay_samples and n < delay_samples + slope_in_samples:
e@0 449 outp[n] = (1. - v) - (1. - v) / slope_in_samples * (n - delay_samples)
e@0 450 elif n >= delay_samples + slope_in_samples and n < len(outp) - delay_samples - slope_out_samples:
e@0 451 outp[n] = 0
e@0 452 elif n >= len(outp) - delay_samples - slope_out_samples and n < len(outp) - delay_samples:
e@0 453 outp[n] = (1. - v) / slope_out_samples * (n - len(outp) + delay_samples + slope_out_samples)
e@0 454 if outp[n] < 0:
e@0 455 print(n)
e@0 456 break
e@0 457 elif n >= len(outp) - delay_samples:
e@0 458 outp[n] = 1.0 - v
e@0 459
e@0 460 outp += v
e@0 461
e@0 462 return outp * x
e@0 463
e@0 464
e@0 465 def get_background(
e@0 466 fname,
e@0 467 duration,
e@0 468 ft=0.5,
e@0 469 ):
e@0 470 print(fname)
e@0 471 bg, sr = librosa.load(fname)
e@0 472 f_s = int(ft * sr)
e@0 473 y = bg
e@0 474 z = np.zeros((duration,))
e@0 475 if len(y) < len(z):
e@0 476 y = fade(y, ft, ft, sr)
e@0 477 for n in range(0, len(z) - len(y), len(y) - f_s):
e@0 478 z[n:n + len(y)] += y
e@0 479 n += len(y) - f_s
e@0 480 if len(y) > len(z[n:]):
e@0 481 z[n:] += y[:len(z[n:])]
e@0 482 else:
e@0 483 z[n:n + len(y)] += y
e@0 484
e@0 485 z = fade(z, ft, ft, sr=sr)
e@0 486
e@0 487 elif len(y) > len(z):
e@0 488 z += fade(y[0:len(z)], ft, ft, sr=sr)
e@0 489 return z
e@0 490
e@0 491
e@0 492 def compose_bg_scene(bgs, background_changes, D, delay=3*44100):
e@0 493 z = np.zeros((2,D))
e@0 494 for n in range(len(background_changes)):
e@0 495 bg_choice = background_changes[n][1]
e@0 496 start = background_changes[n][0]
e@0 497 fname = bgs[bg_choice]
e@0 498 if n < len(background_changes) - 1:
e@0 499 duration = background_changes[n + 1][0] - background_changes[n][0]
e@0 500 else:
e@0 501 duration = D - background_changes[n][0]
e@0 502
e@0 503 y = get_background(fname, duration)
e@0 504 z[0,start:start + len(y)] = y
e@0 505 z[1, start:start + len(y)] = y
e@0 506 #z = fade(z, 1., 1.)
e@0 507 return z
e@0 508
e@0 509
e@0 510 class Director():
e@0 511 def __init__(self, script, sound_dir, speech_dir):
e@0 512 """
e@0 513 Gets a list of script
e@0 514
e@0 515 :param sound_dir: directory of sound files
e@0 516 :param speech_dir: directory of speech files
e@0 517 :param script: the script
e@0 518 """
e@0 519
e@0 520 # Gets character definitions
e@0 521
e@0 522 ## TODO: Change this to also have accents
e@0 523
e@0 524 self.voice_params = {}
e@0 525 self.scene_params = {}
e@0 526 self.bg_params = {}
e@0 527
e@0 528 # This holds the fxive sound engine if available
e@0 529 self.fxive = None
e@0 530
e@0 531 global FFV, FMV
e@0 532 for d in script['definitions']:
e@0 533 if d['type'] == 'scene_definition':
e@0 534 number = int(d['number'])
e@0 535 tags = d['tags']
e@0 536 filename = d['filename']
e@0 537
e@0 538 # If it starts with fxive: then get the preset from fxive
e@0 539 if 'fxive:' == filename[:6]:
e@0 540 print("Fetching sample from fxive...")
e@0 541 if self.fxive is not None:
e@0 542 self.bg_params[number] = self.fxive.get_sfx(filename[6:])
e@0 543 else:
e@0 544 self.fxive = FXive(sfx_path=os.path.join(sound_dir, 'sfx.xls'))
e@0 545 self.bg_params[number] = self.fxive.get_sfx(filename[6:])
e@0 546 else:
e@0 547 self.bg_params[number] = filename
e@0 548
e@0 549 if 'none' in tags:
e@0 550 self.scene_params[number] = []
e@0 551 else:
e@0 552 self.scene_params[number] = tags
e@0 553
e@0 554 if d['type'] == 'cast_definition':
e@0 555 # print("-----------------")
e@0 556 name = d['name']
e@0 557 gender = random.sample(d['gender'], 1)[0]
e@0 558 panning = random.sample(d['panning'], 1)[0]
e@0 559
e@0 560 if panning == 'left':
e@0 561 panning = 0.01
e@0 562 elif panning == 'right':
e@0 563 panning = 0.99
e@0 564 elif panning in ['center', 'centre']:
e@0 565 panning = 0.5
e@0 566 #print(gender, panning)
e@0 567 if gender == 'female':
e@0 568 # Choose a random female voice
e@0 569 voice = random.sample(female_voices, 1)[0]
e@0 570
e@0 571 if FIXED_VOICES:
e@0 572 voice = fixed_female_voice[FFV]
e@0 573 FFV += 1
e@0 574 else:
e@0 575 # Choose a random male voice
e@0 576 voice = random.sample(male_voices, 1)[0]
e@0 577
e@0 578 if FIXED_VOICES:
e@0 579 voice = fixed_male_voice[FMV]
e@0 580 FMV += 1
e@0 581
e@0 582 self.voice_params[name] = (voice, panning)
e@0 583
e@0 584 # if character_panning == 0.5:
e@0 585 # character_panning = 0.1
e@0 586 # elif character_panning == 0.1:
e@0 587 # character_panning = 0.9
e@0 588 # elif character_panning == 0.9:
e@0 589 # character_panning = 0.1
e@0 590
e@0 591
e@0 592 if self.fxive is not None:
e@0 593 self.fxive.close()
e@0 594
e@0 595 self.script = script
e@0 596 self.sound_dir = sound_dir
e@0 597 self.speech_dir = speech_dir
e@0 598
e@0 599 self.musicmanager = MusicManager(sound_dir)
e@0 600 self.pausemanager = PauseManager()
e@0 601 self.speechmanager = SpeechManager(speech_dir, self.voice_params)
e@0 602 self.sfxmanager = SoundManager(sound_dir)
e@0 603
e@0 604 def get_voice_params(self, name):
e@0 605 return self.voice_params[name]
e@0 606
e@0 607 def generate_multitrack(self):
e@0 608 # Shift by 4 seconds
e@0 609 D = 0
e@0 610 P = []
e@0 611 track_names = []
e@0 612
e@0 613 # print(self.script['script'])
e@0 614
e@0 615 current_scene = 1
e@0 616 current_reverb_tags = ""
e@0 617
e@0 618 scene_changes = []
e@0 619
e@0 620 # Create a program of scripts
e@0 621 for s in self.script['script']:
e@0 622 if s['type'] == 'music':
e@0 623 name = 'music'
e@0 624 audio = self.musicmanager.retrieve_music(s)
e@0 625 elif s['type'] == 'sfx':
e@0 626 name = s['name'].lower()
e@0 627 audio = self.sfxmanager.retrieve_sfx(s)
e@0 628 elif s['type'] == 'scene_change':
e@0 629 current_scene = int(s['number'])
e@0 630 #print(current_scene)
e@0 631 #print(self.scene_params)
e@0 632 current_reverb_tags = self.scene_params[current_scene]
e@0 633
e@0 634 print("Changed to scene {} with reverb tags: {}".format(current_scene, current_reverb_tags))
e@0 635 scene_changes.append((D, current_scene))
e@0 636 continue
e@0 637 elif s['type'] == 'pause':
e@0 638 name = 'pause'
e@0 639 audio = self.pausemanager.retrieve_pause(s)
e@0 640 elif s['type'] == 'cast_line':
e@0 641 print(s)
e@0 642 name = s['name'].lower()
e@0 643 audio = self.speechmanager.retrieve_speech(s)
e@0 644 if len(current_reverb_tags) > 0:
e@0 645 print("Applying reverberation with tags: {}".format(current_reverb_tags))
e@0 646 print(audio.shape)
e@0 647 if s['name'] != 'Narrator':
e@0 648 audio = get_reverb_from_tags(audio[0,:], audio[1,:], current_reverb_tags)
e@0 649
e@0 650 if name not in track_names:
e@0 651 track_names.append(name)
e@0 652 D += audio.shape[1]
e@0 653 P.append((name,audio))
e@0 654
e@0 655 multitrack = {t: np.zeros((2, D)) for t in track_names}
e@0 656
e@0 657 print("Composing bg scene")
e@0 658 multitrack['background'] = compose_bg_scene(self.bg_params, scene_changes, D)
e@0 659
e@0 660 idx = 0
e@0 661 for p in P:
e@0 662 multitrack[p[0]][:, idx:idx+p[1].shape[1]] = p[1]
e@0 663 idx += p[1].shape[1]
e@0 664
e@0 665 return multitrack
e@0 666
e@0 667
e@0 668
e@0 669 class Generator():
e@0 670 def __init__(self):
e@0 671 pass
e@0 672
e@0 673 def generate(self):
e@0 674 with open('../data/scripts/The Mystery Of Spooky Hill.txt') as f:
e@0 675 return f.read()
e@0 676
e@0 677
e@0 678 class PauseManager():
e@0 679 def __init__(self):
e@0 680 """
e@0 681 Manages pauses
e@0 682 """
e@0 683
e@0 684 def retrieve_pause(self, input_):
e@0 685 duration_str = input_['duration']
e@0 686 if duration_str == 'long':
e@0 687 duration = 3.0
e@0 688 elif duration_str == 'short':
e@0 689 duration = 1.0
e@0 690
e@0 691 audio = np.zeros((2, int(duration*44100)))
e@0 692 return audio
e@0 693
e@0 694 class SpeechManager():
e@0 695 def __init__(self, speech_folder, voice_params):
e@0 696 """
e@0 697
e@0 698 :param speech_folder: the folder the speech .mp3s are
e@0 699 """
e@0 700
e@0 701 self.voice_params = voice_params
e@0 702 self.speech_folder = speech_folder
e@0 703 try:
e@0 704 self.transcriptions = pd.read_excel(os.path.join(speech_folder ,'transcript.xls'))
e@0 705 except:
e@0 706 # If the file does not exist
e@0 707 self.transcriptions = None
e@0 708
e@0 709 print('Transcription file:' + str(os.path.join(speech_folder ,'transcript.xls')))
e@0 710 print('Transcriptions:' + str(self.transcriptions))
e@0 711
e@0 712 def retrieve_speech(self, input_):
e@0 713 # print(input_)
e@0 714 cast_member = input_['name']
e@0 715 # print(self.voice_params)
e@0 716 cast_voice = self.voice_params[cast_member][0] # 0th element is voice
e@0 717 cast_panning = self.voice_params[cast_member][1] #1th element is panning
e@0 718
e@0 719 cast_line = input_['line']
e@0 720
e@0 721 can_find_entry = False
e@0 722
e@0 723
e@0 724 # If the file does not exist
e@0 725 cast_lines_df = self.transcriptions[self.transcriptions['cast'].map(lambda x: x.lower()) == cast_member.lower()]
e@0 726 similarities = {}
e@0 727 for n in cast_lines_df.index:
e@0 728 similarities[n] = difflib.SequenceMatcher(None, cast_line, cast_lines_df['line'].loc[n]).ratio()
e@0 729
e@0 730 # Most similar entry location
e@0 731 chosen_entry = max(similarities, key=lambda x: similarities[x])
e@0 732 chosen_file = cast_lines_df['filename'].loc[chosen_entry]
e@0 733 chosen_line = cast_lines_df['line'].loc[chosen_entry]
e@0 734
e@0 735 if similar(cast_line, chosen_line):
e@0 736 can_find_entry = True
e@0 737
e@0 738 chosen_file_path = os.path.join(self.speech_folder, chosen_file)
e@0 739 print("Retrieving: " + chosen_file_path)
e@0 740
e@0 741 if os.path.exists(chosen_file_path):
e@0 742 audio, sr = librosa.core.load(chosen_file_path, sr=44100, mono=False)
e@0 743 #print("panning: {}".format(cast_panning))
e@0 744 audio[0,:] *= cast_panning
e@0 745 audio[1,:] *= (1-cast_panning)
e@0 746 else:
e@0 747 can_find_entry = False
e@0 748
e@0 749 if not can_find_entry:
e@0 750 # 1. Generate line
e@0 751 audio, sr = generate_speech_with_festival(cast_voice, cast_panning, cast_line, sr=44100)
e@0 752 # print("panning: {}".format(cast_panning))
e@0 753 # audio[0,:] *= cast_panning
e@0 754 # audio[1,:] *= (1-cast_panning)
e@0 755
e@0 756
e@0 757
e@0 758 # If the line is too disimilar, synthesize it, else use the chosen line
e@0 759 return audio
e@0 760
e@0 761
e@0 762 class SoundManager():
e@0 763 def __init__(self, sound_folder):
e@0 764 """
e@0 765
e@0 766 :param sound_folder: the folder the music .mp3s are
e@0 767 """
e@0 768
e@0 769
e@0 770 self.sound_folder = sound_folder
e@0 771 self.sound_file_names = [f.split('/')[-1] for f in glob.glob(sound_folder + '*.mp3')]
e@0 772
e@0 773 # If the directory is empty, return.
e@0 774 if len(self.sound_file_names) == 0:
e@0 775 return
e@0 776 # Lookup strings
e@0 777 strings = []
e@0 778 for f in self.sound_file_names:
e@0 779 strings.append(" ".join(re.findall('[A-Za-z]+', f)).lower())
e@0 780
e@0 781 # Sanitize strings, remove the most common substring
e@0 782
e@0 783 # Find most common substring
e@0 784 string1 = strings[0]
e@0 785 for n in range(1, len(strings)):
e@0 786 string2 = strings[n]
e@0 787 match = difflib.SequenceMatcher(None, string1, string2).find_longest_match(0, len(string1), 0, len(string2))
e@0 788 string1 = string2[match.b:match.b + match.size]
e@0 789
e@0 790 # Remove most common substring
e@0 791 ## TODO: Check here please, should we remove it?
e@0 792
e@0 793 # strings = [s.replace(string1, '') for s in strings]
e@0 794 self.lookup = strings
e@0 795
e@0 796 def retrieve_sfx(self, input_):
e@0 797 """
e@0 798
e@0 799 :param query: dictionary object from parser
e@0 800 :return: audio matrix containing audio file
e@0 801 """
e@0 802
e@0 803 query = input_['name'].lower()
e@0 804 # Lematize words before checking for similarity
e@0 805 stemmer = porter.PorterStemmer()
e@0 806
e@0 807 qwords = [stemmer.stem(q).lower() for q in query.split()]
e@0 808 similarities = []
e@0 809
e@0 810 # If the words in the query are available in the words in the filename, then increase by 1. Finally,
e@0 811 # divide by the total number of words (Jaccard similarity?)
e@0 812
e@0 813 for s in self.lookup:
e@0 814
e@0 815 words = [stemmer.stem(w).lower() for w in s.split()]
e@0 816 similarities.append(0.)
e@0 817
e@0 818 for qw in qwords:
e@0 819 for w in words:
e@0 820 similarities[-1] += difflib.SequenceMatcher(None, qw, w).ratio()
e@0 821
e@0 822 similarities[-1]/=float(len(words))
e@0 823
e@0 824 # This is argmax
e@0 825 chosen = [n for n in range(len(similarities)) if similarities[n] == max(similarities)][0]
e@0 826 chosen_fname = self.sound_folder + self.sound_file_names[chosen]
e@0 827 audio = librosa.core.load(chosen_fname, sr=44100, mono=False)
e@0 828 return audio[0]
e@0 829
e@0 830
e@0 831 class MusicManager():
e@0 832 def __init__(self, sound_folder):
e@0 833 """
e@0 834
e@0 835 :param sound_folder: the folder the music .mp3s are
e@0 836 """
e@0 837
e@0 838 self.sound_folder = sound_folder
e@0 839 self.sound_file_names = [f.split('/')[-1] for f in glob.glob(sound_folder + '/*.mp3')]
e@0 840
e@0 841 # If the directory is empty, return.
e@0 842 if len(self.sound_file_names) == 0:
e@0 843 return
e@0 844
e@0 845 # Lookup strings
e@0 846 strings = []
e@0 847 for f in self.sound_file_names:
e@0 848 strings.append(" ".join(re.findall('[A-Za-z]+', f)).lower())
e@0 849
e@0 850 # Sanitize strings, remove the most common substring
e@0 851
e@0 852 # Find most common substring
e@0 853 string1 = strings[0]
e@0 854 for n in range(1, len(strings)):
e@0 855 string2 = strings[n]
e@0 856 match = difflib.SequenceMatcher(None, string1, string2).find_longest_match(0, len(string1), 0, len(string2))
e@0 857 string1 = string2[match.b:match.b + match.size]
e@0 858
e@0 859 # Remove most common substring
e@0 860 strings = [s.replace(string1, '') for s in strings]
e@0 861 self.lookup = strings
e@0 862
e@0 863 def retrieve_music(self, input_):
e@0 864 """
e@0 865
e@0 866 :param query: dictionary object from parser
e@0 867 :return: audio matrix containing audio file
e@0 868 """
e@0 869
e@0 870 query = input_['name'].lower() + ' music'
e@0 871
e@0 872 similarities = []
e@0 873
e@0 874 # If the words in the query are available in the words in the filename, then increase by 1. Finally,
e@0 875 # divide by the total number of words (Jaccard similarity?)
e@0 876
e@0 877 for s in self.lookup:
e@0 878 qwords = query.split()
e@0 879 words = s.split()
e@0 880 similarities.append(0.)
e@0 881
e@0 882 for qw in qwords:
e@0 883 if qw in words:
e@0 884 similarities[-1] += 1.
e@0 885
e@0 886 similarities[-1]/=float(len(words))
e@0 887
e@0 888 # This is argmax
e@0 889 chosen = [n for n in range(len(similarities)) if similarities[n] == max(similarities)][0]
e@0 890 chosen_fname = self.sound_folder + self.sound_file_names[chosen]
e@0 891 audio = librosa.core.load(chosen_fname, sr=44100, mono=False)
e@0 892 return audio[0]
e@0 893
e@0 894
e@0 895 # Classes for aiding parsing
e@0 896 class Environment:
e@0 897 def __init__(self, varname, name):
e@0 898 self.name = name
e@0 899 self.varname = varname
e@0 900
e@0 901 def to_json(self):
e@0 902 return {"type": "environment_definition", "name": self.name}
e@0 903
e@0 904
e@0 905 class Sound_Effect:
e@0 906 def __init__(self, varname, name, pos):
e@0 907 self.name = name
e@0 908 self.varname = varname
e@0 909 self.keywords = [kw for kw in name.split()]
e@0 910
e@0 911 # Set the end to pos-1 so the first
e@0 912 # character of the next line won't be ommited
e@0 913
e@0 914 self.pos = (pos, pos - 1)
e@0 915
e@0 916 def to_json(self):
e@0 917 return {
e@0 918 'definition': {
e@0 919 'type': 'sfx_definition',
e@0 920 'name': ' '.join(self.keywords),
e@0 921 'optional': False
e@0 922 },
e@0 923 'playback': {
e@0 924 'type': 'sfx',
e@0 925 'name': ' '.join(self.keywords)
e@0 926 }
e@0 927 }
e@0 928
e@0 929 def add_keywords(self, keywords):
e@0 930 for kw in keywords:
e@0 931 self.keywords.insert(0, kw)
e@0 932
e@0 933 def __str__(self):
e@0 934 return "({} FX)".format(' '.join(self.keywords))
e@0 935
e@0 936 def definition(self):
e@0 937 return ' '.join(self.keywords)
e@0 938
e@0 939 def script(self):
e@0 940 return str(self)
e@0 941
e@0 942
e@0 943 class Character_Line:
e@0 944 def __init__(self, varname, txt, pos_start, pos_end):
e@0 945 self.varname = varname
e@0 946 self.txt = '. '.join([sent.capitalize() for sent in txt.split('\n')])
e@0 947 if self.txt[-1] != '.':
e@0 948 self.txt += '.'
e@0 949
e@0 950 self.character = None
e@0 951 self.pos = (pos_start, pos_end)
e@0 952
e@0 953 def set_character(self, character):
e@0 954 self.character = character
e@0 955
e@0 956 def __str__(self):
e@0 957 return "{}: {}".format(self.character.name, self.txt)
e@0 958
e@0 959 def script(self):
e@0 960 return "[{}] {}".format(self.character.name, self.txt)
e@0 961
e@0 962 def set_pos(self, start, end):
e@0 963 self.pos = (start, end)
e@0 964
e@0 965 def to_json(self):
e@0 966 return {'playback': {"type": "cast_line", "name": self.character.name, "line": self.txt}}
e@0 967
e@0 968
e@0 969 class Character:
e@0 970 def __init__(self, varname, name):
e@0 971 self.name = ' '.join([n.capitalize() for n in name.split()])
e@0 972 self.varname = varname
e@0 973 self.gender = ''
e@0 974 self.age = ''
e@0 975
e@0 976 def set_gender(self, gender):
e@0 977 self.gender = gender
e@0 978
e@0 979 def set_age(self, age):
e@0 980 self.age = age
e@0 981
e@0 982 def definition(self):
e@0 983 str_ = self.name + ' - '
e@0 984 if self.gender == '':
e@0 985 str_ += 'male or female'
e@0 986 else:
e@0 987 str_ += self.gender
e@0 988
e@0 989 return str_
e@0 990
e@0 991 def __str__(self):
e@0 992 return __repr__(self)
e@0 993
e@0 994 def __repr__(self):
e@0 995 return "[{}:{}/{}/{}]".format(self.varname, self.name, self.gender, self.age)
e@0 996
e@0 997 def to_json(self):
e@0 998 json_dict = {"type": "cast_definition", "name": self.name}
e@0 999 if self.gender != '':
e@0 1000 json_dict['gender'] = self.gender
e@0 1001 if self.age != '':
e@0 1002 json_dict['age'] = self.age
e@0 1003
e@0 1004 return json_dict
e@0 1005
e@0 1006
e@0 1007 class KDuration(pg.Keyword):
e@0 1008 grammar = pg.Enum(pg.K('long'), pg.K('short'))
e@0 1009
e@0 1010
e@0 1011 class Pause(pg.Plain):
e@0 1012 grammar = '(', pg.optional(pg.attr('duration', KDuration)), 'pause', ')'
e@0 1013
e@0 1014
e@0 1015 class CastHeader(pg.Plain):
e@0 1016 grammar = 'Cast', pg.optional('List'), ':', pg.endl
e@0 1017
e@0 1018
e@0 1019 class KGender(pg.Keyword):
e@0 1020 grammar = pg.Enum(pg.K('male'), pg.K('female'))
e@0 1021
e@0 1022
e@0 1023 class EGender(pg.List):
e@0 1024 grammar = KGender, pg.optional('or', KGender)
e@0 1025
e@0 1026
e@0 1027 class KPan(pg.Keyword):
e@0 1028 grammar = pg.Enum(pg.K('left'), pg.K('right'), pg.K('center'), pg.K('centre'))
e@0 1029
e@0 1030
e@0 1031 class EPan(pg.List):
e@0 1032 grammar = KPan, pg.optional('or', KPan)
e@0 1033
e@0 1034
e@0 1035 class CastDefinition(pg.Plain):
e@0 1036 grammar = pg.attr('cast_name', re.compile('[A-Za-z0-9 ]+')), \
e@0 1037 re.compile('\-+'), \
e@0 1038 pg.attr('gender', EGender), \
e@0 1039 re.compile('\-+'), \
e@0 1040 'panned', \
e@0 1041 pg.attr('panning', EPan), pg.endl
e@0 1042
e@0 1043
e@0 1044 class Tag(pg.Plain):
e@0 1045 grammar = pg.attr('tag', re.compile(r'[A-Za-z0-9_\-]+'))
e@0 1046
e@0 1047
e@0 1048 class LTag(pg.List):
e@0 1049 grammar = pg.csl(Tag)
e@0 1050
e@0 1051
e@0 1052 class ScenesHeader(pg.Plain):
e@0 1053 grammar = re.compile('Scenes?'), pg.optional('List'), ':', pg.endl
e@0 1054
e@0 1055
e@0 1056 class ScenesDefinition(pg.Plain):
e@0 1057 grammar = pg.attr('number', re.compile('[A-Za-z0-9]+')), \
e@0 1058 re.compile('\-+'), pg.attr('name', re.compile('[A-Za-z0-9]+')), \
e@0 1059 re.compile('\-+'), pg.attr('filename', re.compile('[A-Za-z0-9_\:]+(\.(mp3|wav))?')), \
e@0 1060 re.compile('\-+'), pg.attr('tags', LTag), pg.endl
e@0 1061
e@0 1062
e@0 1063 class ScriptHeader(pg.Plain):
e@0 1064 grammar = 'Script', ':', pg.endl
e@0 1065
e@0 1066
e@0 1067 class SceneCommence(pg.Plain):
e@0 1068 grammar = re.compile('\-+'), 'Scene', pg.attr('scene', re.compile('[A-Za-z0-9]+')), pg.optional(
e@0 1069 re.compile('\-+')), pg.endl;
e@0 1070
e@0 1071
e@0 1072 class CastLine(pg.Plain):
e@0 1073 grammar = '[', pg.attr('cast_name', re.compile('[A-Za-z0-9 ]+')), ']', pg.attr('line',
e@0 1074 re.compile(r'[A-Za-z0-9\-_.\ \" \'\,\?\:\!]+')),
e@0 1075
e@0 1076
e@0 1077 class Headers(pg.Plain):
e@0 1078 grammar = CastHeader, \
e@0 1079 pg.attr('cast_list', pg.maybe_some(CastDefinition)), \
e@0 1080 pg.optional(ScenesHeader, pg.attr('scene_list', pg.maybe_some(ScenesDefinition))), pg.optional(
e@0 1081 ScriptHeader)
e@0 1082
e@0 1083
e@0 1084 class Script(pg.List):
e@0 1085 grammar = pg.some([Pause, SceneCommence, CastLine])
e@0 1086
e@0 1087
e@0 1088 class ScriptDocument(pg.Plain):
e@0 1089 grammar = pg.attr('headers', Headers), pg.attr('script', Script)
e@0 1090
e@0 1091
e@0 1092 class Parser:
e@0 1093 def __init__(self):
e@0 1094 pass
e@0 1095
e@0 1096 def parse_str(self, str_):
e@0 1097
e@0 1098 obj = pg.parse(str_, ScriptDocument)
e@0 1099 definitions = []
e@0 1100 script = []
e@0 1101
e@0 1102 for cast_def in obj.headers.cast_list:
e@0 1103 name = cast_def.cast_name.strip()
e@0 1104 gender = [str(t) for t in cast_def.gender]
e@0 1105
e@0 1106 panning = [str(t) for t in cast_def.panning]
e@0 1107
e@0 1108 cast_dict = {'type': 'cast_definition', 'name': name, 'gender': gender, 'voice': [], 'panning': panning}
e@0 1109 definitions.append(cast_dict)
e@0 1110
e@0 1111 for scene_def in obj.headers.scene_list:
e@0 1112 name = scene_def.name.strip()
e@0 1113 number = scene_def.number
e@0 1114 filename = scene_def.filename
e@0 1115 tags = [str(t.tag) for t in scene_def.tags]
e@0 1116 scene_dict = {'type': 'scene_definition',
e@0 1117 'scene': name,
e@0 1118 'number': number,
e@0 1119 'filename': scene_def.filename,
e@0 1120 'tags': tags}
e@0 1121 definitions.append(scene_dict)
e@0 1122
e@0 1123 for action in obj.script:
e@0 1124 if isinstance(action, Pause):
e@0 1125 duration = str(action.duration)
e@0 1126 pause_dict = {'type': 'pause', 'duration': duration}
e@0 1127 script.append(pause_dict)
e@0 1128 if isinstance(action, SceneCommence):
e@0 1129 number = str(action.scene).strip()
e@0 1130 scene_dict = {'type': 'scene_change', 'number': number}
e@0 1131 script.append(scene_dict)
e@0 1132 if isinstance(action, CastLine):
e@0 1133 name = str(action.cast_name).strip()
e@0 1134 line = str(action.line)
e@0 1135 cast_line = {'type': 'cast_line', 'name': name, 'line': line}
e@0 1136 script.append(cast_line)
e@0 1137
e@0 1138 return {'definitions': definitions, 'script': script}
e@0 1139
e@0 1140
e@0 1141 class Preprocessor():
e@0 1142 def __init__(self):
e@0 1143 self.parsed_dict = None
e@0 1144 self.regexp_entity = re.compile('(?P<variable>[A-Z][0-9]+)\s+(?P<type>[A-Z][A-Za-z_]*)\s+([0-9]+)\s+([0-9]+)(?P<name>(\s+[A-Za-z\'\"]+)+$)')
e@0 1145 self.regexp_attribute = re.compile('(?P<variable>[A-Z][0-9]+)\s+(?P<type>[A-Z][A-Za-z_]*)\s+(?P<target_variable>[A-Z][0-9]+)\s+(?P<value>[A-Za-z]+)')
e@0 1146 self.regexp_relation = re.compile('(?P<variable>[A-Z][0-9]+)\s+(([A-Za-z0-9_]+:[A-Z0-9]+\s*)+)')
e@0 1147 self.regexp_transitive = re.compile('(?P<arity>\*)\s+(?P<type>[A-Z][A-Za-z_]*)\s+(?P<refersTo1>[A-Z][0-9]+)\s+(?P<refersTo2>[A-Z][0-9]+)\s*')
e@0 1148
e@0 1149 def parse_str(self, text, annot):
e@0 1150 """ takes as input a text and an annotation string """
e@0 1151
e@0 1152 tups = [tup for tup in annot.split('\n') if tup.strip() != '']
e@0 1153
e@0 1154 # Add the narrator
e@0 1155 narrator = Character('_', 'Narrator')
e@0 1156 characters = [narrator]
e@0 1157
e@0 1158 objects = []
e@0 1159 character_lines = []
e@0 1160 sound_effects = []
e@0 1161
e@0 1162 def find_character(var):
e@0 1163 for c in characters:
e@0 1164 if c.varname == var:
e@0 1165 return c
e@0 1166
e@0 1167 def find_character_or_object(var):
e@0 1168 c = find_character(var)
e@0 1169 if c is not None:
e@0 1170 return c
e@0 1171
e@0 1172 def find_character_lines(var):
e@0 1173 for c in character_lines:
e@0 1174 if c.varname == var:
e@0 1175 return c
e@0 1176
e@0 1177 def find_sound_effect(var):
e@0 1178 for c in sound_effects:
e@0 1179 if c.varname == var:
e@0 1180 return c
e@0 1181
e@0 1182 def character_speaks(c):
e@0 1183 for cl in character_lines:
e@0 1184 if isinstance(cl, Character_Line) and cl.character == c:
e@0 1185 return True
e@0 1186 return False
e@0 1187
e@0 1188 for tup in tups:
e@0 1189 # print(tup)
e@0 1190
e@0 1191 groups = self.regexp_entity.findall(tup)
e@0 1192 if len(groups) > 0:
e@0 1193 if groups[0][1] == 'Character':
e@0 1194 # If the entity is a character
e@0 1195 characters.append(
e@0 1196 Character(groups[0][0].strip(), groups[0][4].strip())
e@0 1197 )
e@0 1198 elif groups[0][1] == 'Character_Line':
e@0 1199 character_lines.append(
e@0 1200 Character_Line(groups[0][0].strip(), groups[0][4].strip(), int(groups[0][2]), int(groups[0][3]))
e@0 1201 )
e@0 1202 elif groups[0][1] == 'Sound_Effect':
e@0 1203 sound_effects.append(
e@0 1204 Sound_Effect(groups[0][0].strip(), groups[0][4].strip(), int(groups[0][2]))
e@0 1205 )
e@0 1206 continue
e@0 1207
e@0 1208 for tup in tups:
e@0 1209 # Attributes and relations
e@0 1210 groups = self.regexp_attribute.findall(tup)
e@0 1211 if len(groups) > 0:
e@0 1212 if groups[0][1] == 'Gender':
e@0 1213 # if
e@0 1214 c = find_character(groups[0][2].strip())
e@0 1215 c.set_gender(groups[0][3].strip().lower())
e@0 1216 elif groups[0][1] == 'Age':
e@0 1217 c = find_character(groups[0][2].strip())
e@0 1218 c.set_age(groups[0][3].strip().lower())
e@0 1219
e@0 1220 for tup in tups:
e@0 1221 # Attributes and relations
e@0 1222 groups = self.regexp_relation.findall(tup)
e@0 1223 if len(groups) > 0 and groups[0][1][:4] == 'Says':
e@0 1224 # print(groups)
e@0 1225 refs = groups[0][1].split()[1:]
e@0 1226
e@0 1227 # Store who and whats
e@0 1228 whats = []
e@0 1229 who = None
e@0 1230
e@0 1231 for ref in refs:
e@0 1232 type_, var = ref.split(':')
e@0 1233 if type_[:4] == 'WHAT':
e@0 1234 whats.append(var)
e@0 1235 elif type_[:3] == 'WHO':
e@0 1236 who = find_character(var)
e@0 1237
e@0 1238 # find character lines:
e@0 1239 clines = [find_character_lines(w) for w in whats]
e@0 1240
e@0 1241 # Assign characters
e@0 1242 for cl in clines:
e@0 1243 cl.set_character(who)
e@0 1244 elif len(groups) > 0 and groups[0][1][:12] == 'Sound_Effect':
e@0 1245 sfx = find_sound_effect(groups[0][1][13:].split()[0])
e@0 1246 #print(groups)
e@0 1247
e@0 1248 # Store extra keywords
e@0 1249 keywords = []
e@0 1250
e@0 1251 refs = groups[0][1].split()[1:]
e@0 1252 for ref in refs:
e@0 1253 #print(ref)
e@0 1254 type_, var = ref.split(':')
e@0 1255 if type_[:8] == 'CausedBy':
e@0 1256 cause = find_character_or_object(var)
e@0 1257 if cause != None:
e@0 1258 keywords.append(cause.name)
e@0 1259
e@0 1260 sfx.add_keywords(keywords)
e@0 1261
e@0 1262 # %% Calculate line segments for character lines and narration
e@0 1263
e@0 1264 # Add sound effects to the timeline
e@0 1265 clpos = [cl.pos for cl in character_lines]
e@0 1266 clpos += [sfx.pos for sfx in sound_effects]
e@0 1267 clpos = sorted(clpos, key=lambda x: x[0])
e@0 1268
e@0 1269 # Add first narrator line
e@0 1270 cl = Character_Line('_', text[0:clpos[0][0]], 0, clpos[0][0] - 1)
e@0 1271 cl.set_character(narrator)
e@0 1272 character_lines.append(cl)
e@0 1273
e@0 1274 for n in range(len(clpos) - 1):
e@0 1275 if clpos[n][1] != clpos[n + 1][0] - 1:
e@0 1276 cl = Character_Line('_', text[clpos[n][1] + 1:clpos[n + 1][0]].rstrip(), clpos[n][1] + 1,
e@0 1277 clpos[n + 1][0] - 1)
e@0 1278 cl.set_character(narrator)
e@0 1279 character_lines.append(cl)
e@0 1280
e@0 1281 character_lines += sound_effects
e@0 1282 character_lines = sorted(character_lines, key=lambda x: x.pos[1])
e@0 1283
e@0 1284 # parsed_dict = {'definitions': [c.to_json() for c in characters],
e@0 1285 # 'script': [cl.to_json() for cl in character_lines]}
e@0 1286
e@0 1287 # parsed_dict = {'definitions': [], 'script':[]}
e@0 1288 #
e@0 1289 # print("Cast List:")
e@0 1290 # for c in characters:
e@0 1291 # if character_speaks(c):
e@0 1292 # print(c.definition())
e@0 1293 #
e@0 1294 # print("")
e@0 1295 # print("Sound Effects:")
e@0 1296 #
e@0 1297 # for c in sound_effects:
e@0 1298 # print(c.definition())
e@0 1299 #
e@0 1300 # for cl in character_lines:
e@0 1301 # print(cl.script())
e@0 1302
e@0 1303 # Add definitions for characters
e@0 1304 # for c in characters:
e@0 1305 # if character_speaks(c):
e@0 1306 # parsed_dict['definitions'].append(c.to_json())
e@0 1307 #
e@0 1308 # # Add definitions for sound effects
e@0 1309 # for c in sound_effects:
e@0 1310 # parsed_dict['definitions'].append(c.to_json()['definition'])
e@0 1311 #
e@0 1312 #
e@0 1313 # # Add timeline information
e@0 1314 # for cl in character_lines:
e@0 1315 # parsed_dict['script'].append(cl.to_json()['playback'])
e@0 1316
e@0 1317 str_ = "Cast List:\n"
e@0 1318 for c in characters:
e@0 1319 if character_speaks(c):
e@0 1320 str_ += "{}\n".format(c.definition())
e@0 1321
e@0 1322 str_ += "Sound Effects:\n"
e@0 1323 for c in sound_effects:
e@0 1324 str_ += "{}\n".format(c.definition())
e@0 1325
e@0 1326
e@0 1327 for cl in character_lines:
e@0 1328 str_ += "{}\n".format(cl.script())
e@0 1329
e@0 1330
e@0 1331 return str_