File size: 75,057 Bytes
60c3c32 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 |
#!/usr/bin/env bash
# Set bash to 'debug' mode, it will exit on :
# -e 'error', -u 'undefined variable', -o ... 'error in pipeline', -x 'print commands',
set -e
set -u
set -o pipefail
log() {
local fname=${BASH_SOURCE[1]##*/}
echo -e "$(date '+%Y-%m-%dT%H:%M:%S') (${fname}:${BASH_LINENO[0]}:${FUNCNAME[1]}) $*"
}
min() {
local a b
a=$1
for b in "$@"; do
if [ "${b}" -le "${a}" ]; then
a="${b}"
fi
done
echo "${a}"
}
SECONDS=0
# General configuration
stage=1 # Processes starts from the specified stage.
stop_stage=10000 # Processes is stopped at the specified stage.
skip_stages= # Spicify the stage to be skipped
skip_data_prep=false # Skip data preparation stages.
skip_train=false # Skip training stages.
skip_eval=false # Skip decoding and evaluation stages.
skip_packing=true # Skip the packing stage.
skip_upload_hf=true # Skip uploading to huggingface stage.
eval_valid_set=false # Run decoding for the validation set
ngpu=1 # The number of gpus ("0" uses cpu, otherwise use gpu).
num_nodes=1 # The number of nodes.
nj=32 # The number of parallel jobs.
inference_nj=32 # The number of parallel jobs in decoding.
gpu_inference=false # Whether to perform gpu decoding.
dumpdir=dump # Directory to dump features.
expdir=exp # Directory to save experiments.
python=python3 # Specify python to execute espnet commands.
# Data preparation related
local_data_opts= # The options given to local/data.sh.
post_process_local_data_opts= # The options given to local/data.sh for additional processing in stage 4.
# Speed perturbation related
speed_perturb_factors= # perturbation factors, e.g. "0.9 1.0 1.1" (separated by space).
# Feature extraction related
feats_type=raw # Feature type (raw, raw_copy, fbank_pitch, or extracted).
audio_format=flac # Audio format: wav, flac, wav.ark, flac.ark (only in feats_type=raw).
multi_columns_input_wav_scp=false # Enable multi columns mode for input wav.scp for format_wav_scp.py
multi_columns_output_wav_scp=false # Enable multi columns mode for output wav.scp for format_wav_scp.py
fs=16k # Sampling rate.
min_wav_duration=0.1 # Minimum duration in second.
max_wav_duration=30.5 # Maximum duration in second.
# Tokenization related
token_type=bpe # Tokenization type (char or bpe).
nbpe=30 # The number of BPE vocabulary.
bpemode=unigram # Mode of BPE (unigram or bpe).
oov="<unk>" # Out of vocabulary symbol.
blank="<blank>" # CTC blank symbol
sos="<sos>" # Start of sentence symbol
eos="<eos>" # End of sentence symbol
sop="<sop>" # Start of prev/prompt symbol
bpe_input_sentence_size=100000000 # Size of input sentence for BPE.
bpe_nlsyms= # non-linguistic symbols list, separated by a comma or a file containing 1 symbol per line, for BPE
bpe_char_cover=1.0 # character coverage when modeling BPE
hugging_face_model_name_or_path="" # Hugging Face model or path for hugging_face tokenizer
# Ngram model related
use_ngram=false
ngram_exp=
ngram_num=3
# Language model related
use_lm=true # Use language model for decoding.
lm_tag= # Suffix to the result dir for language model training.
lm_exp= # Specify the directory path for LM experiment.
# If this option is specified, lm_tag is ignored.
lm_stats_dir= # Specify the directory path for LM statistics.
lm_config= # Config for language model training.
lm_args= # Arguments for language model training, e.g., "--max_epoch 10".
# Note that it will overwrite args in lm config.
use_word_lm=false # Whether to use word language model.
num_splits_lm=1 # Number of splitting for lm corpus.
# shellcheck disable=SC2034
word_vocab_size=10000 # Size of word vocabulary.
# S2T model related
s2t_task=s2t
s2t_tag= # Suffix to the result dir for s2t model training.
s2t_exp= # Specify the directory path for s2t experiment.
# If this option is specified, s2t_tag is ignored.
s2t_stats_dir= # Specify the directory path for s2t statistics.
s2t_config= # Config for s2t model training.
s2t_args= # Arguments for s2t model training, e.g., "--max_epoch 10".
# Note that it will overwrite args in s2t config.
feats_normalize=global_mvn # Normalizaton layer type.
num_splits_s2t=1 # Number of splitting for lm corpus.
num_ref=1 # Number of references for training.
# In supervised learning based speech enhancement / separation, it is equivalent to number of speakers.
num_inf= # Number of inferences output by the model
# Note that if it is not specified, it will be the same as num_ref. Otherwise, it will be overwritten.
# In MixIT, number of outputs is larger than that of references.
# Upload model related
hf_repo=
# Decoding related
use_streaming=false # Whether to use streaming decoding
batch_size=1
inference_tag= # Suffix to the result dir for decoding.
inference_config= # Config for decoding.
inference_args= # Arguments for decoding, e.g., "--lm_weight 0.1".
# Note that it will overwrite args in inference config.
inference_lm=valid.loss.ave.pth # Language model path for decoding.
inference_ngram=${ngram_num}gram.bin
inference_s2t_model=valid.acc.ave.pth # S2T model path for decoding.
# e.g.
# inference_s2t_model=train.loss.best.pth
# inference_s2t_model=3epoch.pth
# inference_s2t_model=valid.acc.best.pth
# inference_s2t_model=valid.loss.ave.pth
download_model= # Download a model from Model Zoo and use it for decoding.
# [Task dependent] Set the datadir name created by local/data.sh
train_set= # Name of training set.
valid_set= # Name of validation set used for monitoring/tuning network training.
test_sets= # Names of test sets. Multiple items (e.g., both dev and eval sets) can be specified.
bpe_train_text= # Text file path of bpe training set.
lm_train_text= # Text file path of language model training set.
lm_dev_text= # Text file path of language model development set.
lm_test_text= # Text file path of language model evaluation set.
nlsyms_txt=none # Non-linguistic symbol list if existing.
cleaner=none # Text cleaner.
hyp_cleaner=none # Text cleaner for hypotheses (may be used with external tokenizers)
g2p=none # g2p method (needed if token_type=phn).
lang=noinfo # The language type of corpus.
score_opts= # The options given to sclite scoring
local_score_opts= # The options given to local/score.sh.
s2t_speech_fold_length=800 # fold_length for speech data during S2T training.
s2t_text_fold_length=150 # fold_length for text data during S2T training.
lm_fold_length=150 # fold_length for LM training.
help_message=$(cat << EOF
Usage: $0 --train_set "<train_set_name>" --valid_set "<valid_set_name>" --test_sets "<test_set_names>"
Options:
# General configuration
--stage # Processes starts from the specified stage (default="${stage}").
--stop_stage # Processes is stopped at the specified stage (default="${stop_stage}").
--skip_stages # Spicify the stage to be skipped (default="${skip_stages}").
--skip_data_prep # Skip data preparation stages (default="${skip_data_prep}").
--skip_train # Skip training stages (default="${skip_train}").
--skip_eval # Skip decoding and evaluation stages (default="${skip_eval}").
--skip_packing # Skip the packing stage (default="${skip_packing}").
--skip_upload_hf # Skip uploading to huggingface stage (default="${skip_upload_hf}").
--eval_valid_set # Run decoding for the validation set (default="${eval_valid_set}").
--ngpu # The number of gpus ("0" uses cpu, otherwise use gpu, default="${ngpu}").
--num_nodes # The number of nodes (default="${num_nodes}").
--nj # The number of parallel jobs (default="${nj}").
--inference_nj # The number of parallel jobs in decoding (default="${inference_nj}").
--gpu_inference # Whether to perform gpu decoding (default="${gpu_inference}").
--dumpdir # Directory to dump features (default="${dumpdir}").
--expdir # Directory to save experiments (default="${expdir}").
--python # Specify python to execute espnet commands (default="${python}").
# Data preparation related
--local_data_opts # The options given to local/data.sh (default="${local_data_opts}").
# Speed perturbation related
--speed_perturb_factors # speed perturbation factors, e.g. "0.9 1.0 1.1" (separated by space, default="${speed_perturb_factors}").
# Feature extraction related
--feats_type # Feature type (raw, raw_copy, fbank_pitch or extracted, default="${feats_type}").
--audio_format # Audio format: wav, flac, wav.ark, flac.ark (only in feats_type=raw or raw_copy, default="${audio_format}").
--fs # Sampling rate (default="${fs}").
--min_wav_duration # Minimum duration in second (default="${min_wav_duration}").
--max_wav_duration # Maximum duration in second (default="${max_wav_duration}").
# Tokenization related
--token_type # Tokenization type (char or bpe, default="${token_type}").
--nbpe # The number of BPE vocabulary (default="${nbpe}").
--bpemode # Mode of BPE (unigram or bpe, default="${bpemode}").
--oov # Out of vocabulary symbol (default="${oov}").
--blank # CTC blank symbol (default="${blank}").
--sos # sos symbol (default="${sos}").
--eos # eos symbol (default="${eos}").
--sop # sop symbol (default="${sop}").
--bpe_input_sentence_size # Size of input sentence for BPE (default="${bpe_input_sentence_size}").
--bpe_nlsyms # Non-linguistic symbol list for sentencepiece, separated by a comma or a file containing 1 symbol per line . (default="${bpe_nlsyms}").
--bpe_char_cover # Character coverage when modeling BPE (default="${bpe_char_cover}").
# Language model related
--lm_tag # Suffix to the result dir for language model training (default="${lm_tag}").
--lm_exp # Specify the directory path for LM experiment.
# If this option is specified, lm_tag is ignored (default="${lm_exp}").
--lm_stats_dir # Specify the directory path for LM statistics (default="${lm_stats_dir}").
--lm_config # Config for language model training (default="${lm_config}").
--lm_args # Arguments for language model training (default="${lm_args}").
# e.g., --lm_args "--max_epoch 10"
# Note that it will overwrite args in lm config.
--use_word_lm # Whether to use word language model (default="${use_word_lm}").
--word_vocab_size # Size of word vocabulary (default="${word_vocab_size}").
--num_splits_lm # Number of splitting for lm corpus (default="${num_splits_lm}").
# S2T model related
--s2t_tag # Suffix to the result dir for s2t model training (default="${s2t_tag}").
--s2t_exp # Specify the directory path for S2T experiment.
# If this option is specified, s2t_tag is ignored (default="${s2t_exp}").
--s2t_stats_dir # Specify the directory path for S2T statistics (default="${s2t_stats_dir}").
--s2t_config # Config for S2T model training (default="${s2t_config}").
--s2t_args # Arguments for S2T model training (default="${s2t_args}").
# e.g., --s2t_args "--max_epoch 10"
# Note that it will overwrite args in s2t config.
--feats_normalize # Normalizaton layer type (default="${feats_normalize}").
--num_splits_s2t # Number of splitting for lm corpus (default="${num_splits_s2t}").
--num_ref # Number of references for training (default="${num_ref}").
# In supervised learning based speech recognition, it is equivalent to number of speakers.
--num_inf # Number of inference audio generated by the model (default="${num_inf}")
# Note that if it is not specified, it will be the same as num_ref. Otherwise, it will be overwritten.
# Decoding related
--inference_tag # Suffix to the result dir for decoding (default="${inference_tag}").
--inference_config # Config for decoding (default="${inference_config}").
--inference_args # Arguments for decoding (default="${inference_args}").
# e.g., --inference_args "--lm_weight 0.1"
# Note that it will overwrite args in inference config.
--inference_lm # Language model path for decoding (default="${inference_lm}").
--inference_s2t_model # S2T model path for decoding (default="${inference_s2t_model}").
--download_model # Download a model from Model Zoo and use it for decoding (default="${download_model}").
--use_streaming # Whether to use streaming decoding (default="${use_streaming}").
# [Task dependent] Set the datadir name created by local/data.sh
--train_set # Name of training set (required).
--valid_set # Name of validation set used for monitoring/tuning network training (required).
--test_sets # Names of test sets.
# Multiple items (e.g., both dev and eval sets) can be specified (required).
--bpe_train_text # Text file path of bpe training set.
--lm_train_text # Text file path of language model training set.
--lm_dev_text # Text file path of language model development set (default="${lm_dev_text}").
--lm_test_text # Text file path of language model evaluation set (default="${lm_test_text}").
--nlsyms_txt # Non-linguistic symbol list if existing (default="${nlsyms_txt}").
--cleaner # Text cleaner (default="${cleaner}").
--g2p # g2p method (default="${g2p}").
--lang # The language type of corpus (default=${lang}).
--score_opts # The options given to sclite scoring (default="{score_opts}").
--local_score_opts # The options given to local/score.sh (default="{local_score_opts}").
--s2t_speech_fold_length # fold_length for speech data during S2T training (default="${s2t_speech_fold_length}").
--s2t_text_fold_length # fold_length for text data during S2T training (default="${s2t_text_fold_length}").
--lm_fold_length # fold_length for LM training (default="${lm_fold_length}").
EOF
)
log "$0 $*"
# Save command line args for logging (they will be lost after utils/parse_options.sh)
run_args=$(scripts/utils/print_args.sh $0 "$@")
. utils/parse_options.sh
if [ $# -ne 0 ]; then
log "${help_message}"
log "Error: No positional arguments are required."
exit 2
fi
. ./path.sh
. ./cmd.sh
# Check required arguments
if ! "${skip_train}"; then
[ -z "${train_set}" ] && { log "${help_message}"; log "Error: --train_set is required"; exit 2; };
[ -z "${valid_set}" ] && { log "${help_message}"; log "Error: --valid_set is required"; exit 2; };
fi
if ! "${eval_valid_set}"; then
[ -z "${test_sets}" ] && { log "${help_message}"; log "Error: --test_sets is required"; exit 2; };
else
[ -z "${valid_set}" ] && { log "${help_message}"; log "Error: --valid_set is required"; exit 2; };
fi
if [ -n "${train_set}" ] && [ "${train_set}" = "${valid_set}" ]; then
log "Error: train_set and valid_set must be different. --train_set ${train_set} --valid_set ${valid_set}"
exit 1
fi
_test_sets=
for dset in ${test_sets}; do
if [ "${dset}" = "${train_set}" ]; then
log "Error: train_set and test_sets must be different. --train_set ${train_set} --test_sets ${test_sets}"
exit 1
fi
if [ "${dset}" = "${valid_set}" ]; then
log "Info: The valid_set '${valid_set}' is included in the test_sets. '--eval_valid_set true' is set and '${valid_set}' is removed from the test_sets"
eval_valid_set=true
elif [[ " ${_test_sets} " =~ [[:space:]]${dset}[[:space:]] ]]; then
log "Info: ${dset} is duplicated in the test_sets. One is removed"
else
_test_sets+="${dset} "
fi
done
test_sets=${_test_sets}
# Check feature type
if [ "${feats_type}" = raw ]; then
data_feats=${dumpdir}/raw
elif [ "${feats_type}" = raw_copy ]; then
# raw_copy is as same as raw except for skipping the format_wav stage
data_feats=${dumpdir}/raw_copy
elif [ "${feats_type}" = fbank_pitch ]; then
data_feats=${dumpdir}/fbank_pitch
elif [ "${feats_type}" = fbank ]; then
data_feats=${dumpdir}/fbank
elif [ "${feats_type}" == extracted ]; then
data_feats=${dumpdir}/extracted
else
log "${help_message}"
log "Error: not supported: --feats_type ${feats_type}"
exit 2
fi
# Extra files for prev/prompt and ASR CTC
utt_extra_files="text.prev text.ctc"
num_inf=${num_inf:=${num_ref}}
# Preprocessor related
if [ ${num_ref} -eq 1 ]; then
# For single speaker, text file path and name are text
ref_text_files_str="text "
ref_text_names_str="text "
else
# For multiple speakers, text file path and name are text_spk[1-N] and [text, text_spk2, ...]
#TODO(simpleoier): later to support flexibly defined text prefix
ref_text_files_str="text_spk1 "
ref_text_names_str="text "
for n in $(seq 2 ${num_ref}); do
ref_text_files_str+="text_spk${n} "
ref_text_names_str+="text_spk${n} "
done
fi
# shellcheck disable=SC2206
ref_text_files=(${ref_text_files_str// / })
# shellcheck disable=SC2206
ref_text_names=(${ref_text_names_str// / })
[ -z "${bpe_train_text}" ] && bpe_train_text="${data_feats}/org/${train_set}/${ref_text_files[0]}"
# Use the same text as S2T for lm training if not specified.
[ -z "${lm_train_text}" ] && lm_train_text="${data_feats}/org/${train_set}/${ref_text_files[0]}"
# Use the same text as S2T for lm training if not specified.
[ -z "${lm_dev_text}" ] && lm_dev_text="${data_feats}/org/${valid_set}/${ref_text_files[0]}"
if [ -z "${lm_test_text}" ]; then
if [ -z "${test_sets}" ]; then
lm_test_text="${data_feats}/org/${valid_set}/${ref_text_files[0]}"
else
# Use the text of the 1st evaldir if lm_test is not specified
lm_test_text="${data_feats}/${test_sets%% *}/${ref_text_files[0]}"
fi
fi
# Check tokenization type
if [ "${lang}" != noinfo ]; then
token_listdir=data/${lang}_token_list
else
token_listdir=data/token_list
fi
bpedir="${token_listdir}/bpe_${bpemode}${nbpe}"
bpeprefix="${bpedir}"/bpe
bpemodel="${bpeprefix}".model
bpetoken_list="${bpedir}"/tokens.txt
chartoken_list="${token_listdir}"/char/tokens.txt
hugging_face_token_list="${token_listdir}/hugging_face_"${hugging_face_model_name_or_path/\//-}/tokens.txt
# NOTE: keep for future development.
# shellcheck disable=SC2034
wordtoken_list="${token_listdir}"/word/tokens.txt
if [ "${token_type}" = bpe ]; then
token_list="${bpetoken_list}"
elif [ "${token_type}" = char ]; then
token_list="${chartoken_list}"
bpemodel=none
elif [ "${token_type}" = word ]; then
token_list="${wordtoken_list}"
bpemodel=none
elif [ "${token_type}" = whisper_en ]; then # should make token_list an output filepath here
token_list="${token_listdir}"/whisper_en/tokens.txt
bpemodel=whisper_en
hyp_cleaner=${cleaner}
elif [ "${token_type}" = whisper_multilingual ]; then
token_list="${token_listdir}"/whisper_multilingual/tokens.txt
bpemodel=whisper_multilingual
hyp_cleaner=${cleaner}
elif [ "${token_type}" = hugging_face ]; then
token_list="${hugging_face_token_list}"
bpemodel=${hugging_face_model_name_or_path}
else
log "Error: not supported --token_type '${token_type}'"
exit 2
fi
if ${use_word_lm}; then
log "Error: Word LM is not supported yet"
exit 2
else
lm_token_list="${token_list}"
lm_token_type="${token_type}"
fi
# Set tag for naming of model directory
if [ -z "${s2t_tag}" ]; then
if [ -n "${s2t_config}" ]; then
s2t_tag="$(basename "${s2t_config}" .yaml)_${feats_type}"
else
s2t_tag="train_${feats_type}"
fi
if [ "${lang}" != noinfo ]; then
s2t_tag+="_${lang}_${token_type}"
else
s2t_tag+="_${token_type}"
fi
if [ "${token_type}" = bpe ]; then
s2t_tag+="${nbpe}"
fi
if [ "${token_type}" = hugging_face ]; then
s2t_tag+="_"${hugging_face_model_name_or_path/\//-}
fi
# Add overwritten arg's info
if [ -n "${s2t_args}" ]; then
s2t_tag+="$(echo "${s2t_args}" | sed -e "s/--/\_/g" -e "s/[ |=/]//g")"
fi
if [ -n "${speed_perturb_factors}" ]; then
s2t_tag+="_sp"
fi
fi
if [ -z "${lm_tag}" ]; then
if [ -n "${lm_config}" ]; then
lm_tag="$(basename "${lm_config}" .yaml)"
else
lm_tag="train"
fi
if [ "${lang}" != noinfo ]; then
lm_tag+="_${lang}_${lm_token_type}"
else
lm_tag+="_${lm_token_type}"
fi
if [ "${lm_token_type}" = bpe ]; then
lm_tag+="${nbpe}"
fi
# Add overwritten arg's info
if [ -n "${lm_args}" ]; then
lm_tag+="$(echo "${lm_args}" | sed -e "s/--/\_/g" -e "s/[ |=/]//g")"
fi
fi
# The directory used for collect-stats mode
if [ -z "${s2t_stats_dir}" ]; then
if [ "${lang}" != noinfo ]; then
s2t_stats_dir="${expdir}/s2t_stats_${feats_type}_${lang}_${token_type}"
else
s2t_stats_dir="${expdir}/s2t_stats_${feats_type}_${token_type}"
fi
if [ "${token_type}" = bpe ]; then
s2t_stats_dir+="${nbpe}"
fi
if [ "${token_type}" = hugging_face ]; then
s2t_stats_dir+="_"${hugging_face_model_name_or_path/\//-}
fi
if [ -n "${speed_perturb_factors}" ]; then
s2t_stats_dir+="_sp"
fi
fi
if [ -z "${lm_stats_dir}" ]; then
if [ "${lang}" != noinfo ]; then
lm_stats_dir="${expdir}/lm_stats_${lang}_${lm_token_type}"
else
lm_stats_dir="${expdir}/lm_stats_${lm_token_type}"
fi
if [ "${lm_token_type}" = bpe ]; then
lm_stats_dir+="${nbpe}"
fi
fi
# The directory used for training commands
if [ -z "${s2t_exp}" ]; then
s2t_exp="${expdir}/s2t_${s2t_tag}"
fi
if [ -z "${lm_exp}" ]; then
lm_exp="${expdir}/lm_${lm_tag}"
fi
if [ -z "${ngram_exp}" ]; then
ngram_exp="${expdir}/ngram"
fi
if [ -z "${inference_tag}" ]; then
if [ -n "${inference_config}" ]; then
inference_tag="$(basename "${inference_config}" .yaml)"
else
inference_tag=inference
fi
# Add overwritten arg's info
if [ -n "${inference_args}" ]; then
inference_tag+="$(echo "${inference_args}" | sed -e "s/--/\_/g" -e "s/[ |=]//g")"
fi
if "${use_lm}"; then
inference_tag+="_lm_$(basename "${lm_exp}")_$(echo "${inference_lm}" | sed -e "s/\//_/g" -e "s/\.[^.]*$//g")"
fi
if "${use_ngram}"; then
inference_tag+="_ngram_$(basename "${ngram_exp}")_$(echo "${inference_ngram}" | sed -e "s/\//_/g" -e "s/\.[^.]*$//g")"
fi
inference_tag+="_s2t_model_$(echo "${inference_s2t_model}" | sed -e "s/\//_/g" -e "s/\.[^.]*$//g")"
fi
if "${skip_data_prep}"; then
skip_stages+="1 2 3 4 5 "
fi
if "${skip_train}"; then
skip_stages+="2 4 5 6 7 8 9 10 11 "
elif ! "${use_lm}"; then
skip_stages+="6 7 8 "
fi
if ! "${use_ngram}"; then
skip_stages+="9 "
fi
if "${skip_eval}"; then
skip_stages+="12 13 "
fi
if [ "${skip_packing}" = "true" ] || [ -n "${download_model}" ]; then
skip_stages+="14 "
fi
if "${skip_upload_hf}"; then
skip_stages+="15 "
fi
skip_stages=$(echo "${skip_stages}" | tr ' ' '\n' | sort -nu | tr '\n' ' ')
log "Skipped stages: ${skip_stages}"
# ========================== Main stages start from here. ==========================
if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ] && ! [[ " ${skip_stages} " =~ [[:space:]]1[[:space:]] ]]; then
log "Stage 1: Data preparation for data/${train_set}, data/${valid_set}, etc."
# [Task dependent] Need to create data.sh for new corpus
local/data.sh ${local_data_opts}
fi
if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ] && ! [[ " ${skip_stages} " =~ [[:space:]]2[[:space:]] ]]; then
if [ -n "${speed_perturb_factors}" ]; then
log "Stage 2: Speed perturbation: data/${train_set} -> data/${train_set}_sp"
for factor in ${speed_perturb_factors}; do
if python3 -c "assert ${factor} != 1.0" 2>/dev/null; then
scripts/utils/perturb_data_dir_speed.sh \
--utt_extra_files "${utt_extra_files} ${ref_text_files_str}" \
"${factor}" "data/${train_set}" "data/${train_set}_sp${factor}"
_dirs+="data/${train_set}_sp${factor} "
else
# If speed factor is 1, same as the original
_dirs+="data/${train_set} "
fi
done
utils/combine_data.sh \
--extra_files "${utt_extra_files} ${ref_text_files_str}" \
"data/${train_set}_sp" ${_dirs}
else
log "Skip stage 2: Speed perturbation"
fi
fi
if [ -n "${speed_perturb_factors}" ]; then
train_set="${train_set}_sp"
fi
if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ] && ! [[ " ${skip_stages} " =~ [[:space:]]3[[:space:]] ]]; then
if "${skip_train}"; then
if "${eval_valid_set}"; then
_dsets="${valid_set} ${test_sets}"
else
_dsets="${test_sets}"
fi
else
_dsets="${train_set} ${valid_set} ${test_sets}"
fi
if [ "${feats_type}" = raw ]; then
log "Stage 3: Format wav.scp: data/ -> ${data_feats}"
# ====== Recreating "wav.scp" ======
# Kaldi-wav.scp, which can describe the file path with unix-pipe, like "cat /some/path |",
# shouldn't be used in training process.
# "format_wav_scp.sh" dumps such pipe-style-wav to real audio file
# and it can also change the audio-format and sampling rate.
# If nothing is need, then format_wav_scp.sh does nothing:
# i.e. the input file format and rate is same as the output.
for dset in ${_dsets}; do
if [ "${dset}" = "${train_set}" ] || [ "${dset}" = "${valid_set}" ]; then
_suf="/org"
else
_suf=""
fi
utils/copy_data_dir.sh --validate_opts --non-print data/"${dset}" "${data_feats}${_suf}/${dset}"
rm -f ${data_feats}${_suf}/${dset}/{segments,wav.scp,reco2file_and_channel,reco2dur}
# Copy extra text files
for extra_txt in ${utt_extra_files}; do
[ -f data/${dset}/${extra_txt} ] && cp data/${dset}/${extra_txt} ${data_feats}${_suf}/${dset}
done
# Copy reference text files if there is more than 1 reference
if [ ${#ref_text_files[@]} -gt 1 ]; then
# shellcheck disable=SC2068
for ref_txt in ${ref_text_files[@]}; do
[ -f data/${dset}/${ref_txt} ] && cp data/${dset}/${ref_txt} ${data_feats}${_suf}/${dset}
done
fi
_opts=
if [ -e data/"${dset}"/segments ]; then
# "segments" is used for splitting wav files which are written in "wav".scp
# into utterances. The file format of segments:
# <segment_id> <record_id> <start_time> <end_time>
# "e.g. call-861225-A-0050-0065 call-861225-A 5.0 6.5"
# Where the time is written in seconds.
_opts+="--segments data/${dset}/segments "
fi
# shellcheck disable=SC2086
scripts/audio/format_wav_scp.sh --nj "${nj}" --cmd "${train_cmd}" \
--audio-format "${audio_format}" --fs "${fs}" ${_opts} \
--multi-columns-input "${multi_columns_input_wav_scp}" \
--multi-columns-output "${multi_columns_output_wav_scp}" \
"data/${dset}/wav.scp" "${data_feats}${_suf}/${dset}"
echo "${feats_type}" > "${data_feats}${_suf}/${dset}/feats_type"
if "${multi_columns_output_wav_scp}"; then
echo "multi_${audio_format}" > "${data_feats}${_suf}/${dset}/audio_format"
else
echo "${audio_format}" > "${data_feats}${_suf}/${dset}/audio_format"
fi
done
elif [ "${feats_type}" = raw_copy ]; then
# If you guaranteed that the data already satisfy the raw format, you can skip format_wav_scp.py for reduce the overhead
for dset in ${_dsets}; do
if [ -e "data/${dset}/segments" ]; then
log "Error: data/${dset}/segments is existing. Please use --feats_type raw"
exit 1
fi
if [ "${dset}" = "${train_set}" ] || [ "${dset}" = "${valid_set}" ]; then
_suf="/org"
else
_suf=""
fi
utils/copy_data_dir.sh --validate_opts --non-print data/"${dset}" "${data_feats}${_suf}/${dset}"
if [ "${dset}" = "${train_set}" ] || [ "${dset}" = "${valid_set}" ]; then
_suf="/org"
if [ -e "data/${dset}/utt2dur" ]; then
_fs=$(python3 -c "import humanfriendly as h;print(h.parse_size('${fs}'))")
<data/${dset}/utt2dur awk '{ print $1, int($2*'${_fs}'); }' > "${data_feats}${_suf}/${dset}"/utt2num_samples
elif [ -e "data/${dset}/utt2num_samples" ]; then
cp "data/${dset}/utt2num_samples" "${data_feats}${_suf}/${dset}"/utt2num_samples
else
log "Error: data/${dset}/utt2dur or data/${dset}/utt2num_samples must be existing for train_set and valid_set. Please use --feats_type raw. If you'd like to perform this script for evaluation, please give --skip_train true"
exit 1
fi
fi
# Copy extra text files
for extra_txt in ${utt_extra_files}; do
[ -f data/${dset}/${extra_txt} ] && cp data/${dset}/${extra_txt} ${data_feats}${_suf}/${dset}
done
# Copy reference text files if there is more than 1 reference
if [ ${#ref_text_files[@]} -gt 1 ]; then
# shellcheck disable=SC2068
for ref_txt in ${ref_text_files[@]}; do
[ -f data/${dset}/${ref_txt} ] && cp data/${dset}/${ref_txt} ${data_feats}${_suf}/${dset}
done
fi
echo "raw" > "${data_feats}${_suf}/${dset}/feats_type"
if "${multi_columns_input_wav_scp}"; then
echo "multi_${audio_format}" > "${data_feats}${_suf}/${dset}/audio_format"
else
echo "${audio_format}" > "${data_feats}${_suf}/${dset}/audio_format"
fi
done
elif [ "${feats_type}" = fbank_pitch ]; then
log "[Require Kaldi] Stage 3: ${feats_type} extract: data/ -> ${data_feats}"
for dset in ${_dsets}; do
if [ "${dset}" = "${train_set}" ] || [ "${dset}" = "${valid_set}" ]; then
_suf="/org"
else
_suf=""
fi
# 1. Copy datadir
utils/copy_data_dir.sh --validate_opts --non-print data/"${dset}" "${data_feats}${_suf}/${dset}"
# Copy extra text files
for extra_txt in ${utt_extra_files}; do
[ -f data/${dset}/${extra_txt} ] && cp data/${dset}/${extra_txt} ${data_feats}${_suf}/${dset}
done
# Copy reference text files if there is more than 1 reference
if [ ${#ref_text_files[@]} -gt 1 ]; then
# shellcheck disable=SC2068
for ref_txt in ${ref_text_files[@]}; do
[ -f data/${dset}/${ref_txt} ] && cp data/${dset}/${ref_txt} ${data_feats}${_suf}/${dset}
done
fi
# 2. Feature extract
_nj=$(min "${nj}" "$(<"${data_feats}${_suf}/${dset}/utt2spk" wc -l)")
steps/make_fbank_pitch.sh --nj "${_nj}" --cmd "${train_cmd}" "${data_feats}${_suf}/${dset}"
utils/fix_data_dir.sh "${data_feats}${_suf}/${dset}"
# 3. Derive the the frame length and feature dimension
scripts/feats/feat_to_shape.sh --nj "${_nj}" --cmd "${train_cmd}" \
"${data_feats}${_suf}/${dset}/feats.scp" "${data_feats}${_suf}/${dset}/feats_shape"
# 4. Write feats_dim
head -n 1 "${data_feats}${_suf}/${dset}/feats_shape" | awk '{ print $2 }' \
| cut -d, -f2 > ${data_feats}${_suf}/${dset}/feats_dim
# 5. Write feats_type
echo "${feats_type}" > "${data_feats}${_suf}/${dset}/feats_type"
done
elif [ "${feats_type}" = fbank ]; then
log "Stage 3: ${feats_type} extract: data/ -> ${data_feats}"
log "${feats_type} is not supported yet."
exit 1
elif [ "${feats_type}" = extracted ]; then
log "Stage 3: ${feats_type} extract: data/ -> ${data_feats}"
# Assumming you don't have wav.scp, but feats.scp is created by local/data.sh instead.
for dset in ${_dsets}; do
if [ "${dset}" = "${train_set}" ] || [ "${dset}" = "${valid_set}" ]; then
_suf="/org"
else
_suf=""
fi
# Generate dummy wav.scp to avoid error by copy_data_dir.sh
if [ ! -f data/"${dset}"/wav.scp ]; then
if [ ! -f data/"${dset}"/segments ]; then
<data/"${dset}"/feats.scp awk ' { print($1,"<DUMMY>") }' > data/"${dset}"/wav.scp
else
<data/"${dset}"/segments awk ' { print($2,"<DUMMY>") }' > data/"${dset}"/wav.scp
fi
fi
utils/copy_data_dir.sh --validate_opts --non-print data/"${dset}" "${data_feats}${_suf}/${dset}"
# Copy extra text files
for extra_txt in ${utt_extra_files}; do
[ -f data/${dset}/${extra_txt} ] && cp data/${dset}/${extra_txt} ${data_feats}${_suf}/${dset}
done
# Copy reference text files if there is more than 1 reference
# shellcheck disable=SC2068
if [ ${#ref_text_files[@]} -gt 1 ]; then
for ref_txt in ${ref_text_files[@]}; do
[ -f data/${dset}/${ref_txt} ] && cp data/${dset}/${ref_txt} ${data_feats}${_suf}/${dset}
done
fi
# Derive the the frame length and feature dimension
_nj=$(min "${nj}" "$(<"${data_feats}${_suf}/${dset}/utt2spk" wc -l)")
scripts/feats/feat_to_shape.sh --nj "${_nj}" --cmd "${train_cmd}" \
"${data_feats}${_suf}/${dset}/feats.scp" "${data_feats}${_suf}/${dset}/feats_shape"
pyscripts/feats/feat-to-shape.py "scp:head -n 1 ${data_feats}${_suf}/${dset}/feats.scp |" - | \
awk '{ print $2 }' | cut -d, -f2 > "${data_feats}${_suf}/${dset}/feats_dim"
echo "${feats_type}" > "${data_feats}${_suf}/${dset}/feats_type"
done
else
log "Error: not supported: --feats_type ${feats_type}"
exit 2
fi
fi
if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ] && ! [[ " ${skip_stages} " =~ [[:space:]]4[[:space:]] ]]; then
log "Stage 4: Remove long/short data: ${data_feats}/org -> ${data_feats}"
# NOTE(kamo): Not applying to test_sets to keep original data
for dset in "${train_set}" "${valid_set}"; do
# Copy data dir
utils/copy_data_dir.sh --validate_opts --non-print "${data_feats}/org/${dset}" "${data_feats}/${dset}"
cp "${data_feats}/org/${dset}/feats_type" "${data_feats}/${dset}/feats_type"
# Remove short utterances
_feats_type="$(<${data_feats}/${dset}/feats_type)"
if [ "${_feats_type}" = raw ]; then
_fs=$(python3 -c "import humanfriendly as h;print(h.parse_size('${fs}'))")
_min_length=$(python3 -c "print(int(${min_wav_duration} * ${_fs}))")
_max_length=$(python3 -c "print(int(${max_wav_duration} * ${_fs}))")
# utt2num_samples is created by format_wav_scp.sh
<"${data_feats}/org/${dset}/utt2num_samples" \
awk -v min_length="${_min_length}" -v max_length="${_max_length}" \
'{ if ($2 > min_length && $2 < max_length ) print $0; }' \
>"${data_feats}/${dset}/utt2num_samples"
<"${data_feats}/org/${dset}/wav.scp" \
utils/filter_scp.pl "${data_feats}/${dset}/utt2num_samples" \
>"${data_feats}/${dset}/wav.scp"
else
# Get frame shift in ms from conf/fbank.conf
_frame_shift=
if [ -f conf/fbank.conf ] && [ "$(<conf/fbank.conf grep -c frame-shift)" -gt 0 ]; then
# Assume using conf/fbank.conf for feature extraction
_frame_shift="$(<conf/fbank.conf grep frame-shift | sed -e 's/[-a-z =]*\([0-9]*\)/\1/g')"
fi
if [ -z "${_frame_shift}" ]; then
# If not existing, use the default number in Kaldi (=10ms).
# If you are using different number, you have to change the following value manually.
_frame_shift=10
fi
_min_length=$(python3 -c "print(int(${min_wav_duration} / ${_frame_shift} * 1000))")
_max_length=$(python3 -c "print(int(${max_wav_duration} / ${_frame_shift} * 1000))")
cp "${data_feats}/org/${dset}/feats_dim" "${data_feats}/${dset}/feats_dim"
<"${data_feats}/org/${dset}/feats_shape" awk -F, ' { print $1 } ' \
| awk -v min_length="${_min_length}" -v max_length="${_max_length}" \
'{ if ($2 > min_length && $2 < max_length) print $0; }' \
>"${data_feats}/${dset}/feats_shape"
<"${data_feats}/org/${dset}/feats.scp" \
utils/filter_scp.pl "${data_feats}/${dset}/feats_shape" \
>"${data_feats}/${dset}/feats.scp"
fi
# Remove empty text
# shellcheck disable=SC2068
for extra_txt in ${utt_extra_files}; do
<"${data_feats}/org/${dset}/${extra_txt}" \
awk ' { if( NF != 1 ) print $0; } ' >"${data_feats}/${dset}/${extra_txt}"
done
for ref_txt in ${ref_text_files[@]}; do
<"${data_feats}/org/${dset}/${ref_txt}" \
awk ' { if( NF != 1 ) print $0; } ' >"${data_feats}/${dset}/${ref_txt}"
done
# fix_data_dir.sh leaves only utts which exist in all files
utils/fix_data_dir.sh \
--utt_extra_files "${utt_extra_files} ${ref_text_files_str}" \
"${data_feats}/${dset}"
done
if [ -n "${post_process_local_data_opts}" ]; then
# Do any additional local data post-processing here
local/data.sh ${post_process_local_data_opts} --s2t_data_dir "${data_feats}/${train_set}"
fi
# shellcheck disable=SC2002,SC2068,SC2005
for lm_txt in ${lm_train_text[@]}; do
suffix=$(echo "$(basename ${lm_txt})" | sed 's/text//')
<${lm_txt} awk -v suffix=${suffix} ' { if( NF != 1 ) {$1=$1 suffix; print $0; }} '
done > "${data_feats}/lm_train.txt"
fi
if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ] && ! [[ " ${skip_stages} " =~ [[:space:]]5[[:space:]] ]]; then
if [ "${token_type}" = bpe ]; then
log "Stage 5: Generate token_list from ${bpe_train_text} using BPE"
mkdir -p "${bpedir}"
# shellcheck disable=SC2002
cat ${bpe_train_text} | cut -f 2- -d" " > "${bpedir}"/train.txt
if [ -n "${bpe_nlsyms}" ]; then
if test -f "${bpe_nlsyms}"; then
bpe_nlsyms_list=$(awk '{print $1}' ${bpe_nlsyms} | paste -s -d, -)
_opts_spm="--user_defined_symbols=${bpe_nlsyms_list}"
else
_opts_spm="--user_defined_symbols=${bpe_nlsyms}"
fi
else
_opts_spm=""
fi
spm_train \
--input="${bpedir}"/train.txt \
--vocab_size="${nbpe}" \
--model_type="${bpemode}" \
--model_prefix="${bpeprefix}" \
--character_coverage=${bpe_char_cover} \
--input_sentence_size="${bpe_input_sentence_size}" \
${_opts_spm}
{
echo "${blank}"
echo "${oov}"
# Remove <unk>, <s>, </s> from the vocabulary
<"${bpeprefix}".vocab awk '{ if( NR != 1 && NR != 2 && NR != 3 ){ print $1; } }'
echo "${sos}"
echo "${eos}"
echo "${sop}"
} > "${token_list}"
elif [ "${token_type}" = char ] || [ "${token_type}" = word ]; then
log "Stage 5: Generate character level token_list from ${lm_train_text}"
_opts="--non_linguistic_symbols ${nlsyms_txt}"
# The first symbol in token_list must be "<blank>" and the last must be also sos/eos:
# 0 is reserved for CTC-blank for ASR and also used as ignore-index in the other task
${python} -m espnet2.bin.tokenize_text \
--token_type "${token_type}" \
--input "${data_feats}/lm_train.txt" --output "${token_list}" ${_opts} \
--field 2- \
--cleaner "${cleaner}" \
--g2p "${g2p}" \
--write_vocabulary true \
--add_symbol "${blank}:0" \
--add_symbol "${oov}:1" \
--add_symbol "${sop}:-1" \
--add_symbol "${eos}:-2" \
--add_symbol "${sos}:-3"
elif grep -q "whisper" <<< ${token_type}; then
log "Stage 5: Generate whisper token_list from ${token_type} tokenizer"
# The first symbol in token_list must be "<blank>" and the last must be also sos/eos:
# 0 is reserved for CTC-blank for ASR and also used as ignore-index in the other task
echo ${token_list}
${python} -m espnet2.bin.whisper_export_vocabulary \
--whisper_model "${token_type}" \
--output "${token_list}"
elif [ "${token_type}" = hugging_face ]; then
log "Stage 5: Generate hugging_face token_list from ${hugging_face_model_name_or_path}"
# The first symbol in token_list must be "<blank>" and the last must be also sos/eos:
# 0 is reserved for CTC-blank for ASR and also used as ignore-index in the other task
${python} -m espnet2.bin.hugging_face_export_vocabulary \
--model_name_or_path "${hugging_face_model_name_or_path}" \
--output "${token_list}"
else
log "Error: not supported --token_type '${token_type}'"
exit 2
fi
# Create word-list for word-LM training
if ${use_word_lm} && [ "${token_type}" != word ]; then
log "Generate word level token_list from ${data_feats}/lm_train.txt"
${python} -m espnet2.bin.tokenize_text \
--token_type word \
--input "${data_feats}/lm_train.txt" --output "${lm_token_list}" \
--field 2- \
--cleaner "${cleaner}" \
--g2p "${g2p}" \
--write_vocabulary true \
--vocabulary_size "${word_vocab_size}" \
--add_symbol "${blank}:0" \
--add_symbol "${oov}:1" \
--add_symbol "${sop}:-1" \
--add_symbol "${eos}:-2" \
--add_symbol "${sos}:-3"
fi
fi
# ========================== Data preparation is done here. ==========================
if [ ${stage} -le 6 ] && [ ${stop_stage} -ge 6 ] && ! [[ " ${skip_stages} " =~ [[:space:]]6[[:space:]] ]]; then
log "Stage 6: LM collect stats: train_set=${data_feats}/lm_train.txt, dev_set=${lm_dev_text}"
_opts=
if [ -n "${lm_config}" ]; then
# To generate the config file: e.g.
# % python3 -m espnet2.bin.lm_train --print_config --optim adam
_opts+="--config ${lm_config} "
fi
# 1. Split the key file
_logdir="${lm_stats_dir}/logdir"
mkdir -p "${_logdir}"
# Get the minimum number among ${nj} and the number lines of input files
_nj=$(min "${nj}" "$(<${data_feats}/lm_train.txt wc -l)" "$(<${lm_dev_text} wc -l)")
key_file="${data_feats}/lm_train.txt"
split_scps=""
for n in $(seq ${_nj}); do
split_scps+=" ${_logdir}/train.${n}.scp"
done
# shellcheck disable=SC2086
utils/split_scp.pl "${key_file}" ${split_scps}
key_file="${lm_dev_text}"
split_scps=""
for n in $(seq ${_nj}); do
split_scps+=" ${_logdir}/dev.${n}.scp"
done
# shellcheck disable=SC2086
utils/split_scp.pl "${key_file}" ${split_scps}
# 2. Generate run.sh
log "Generate '${lm_stats_dir}/run.sh'. You can resume the process from stage 6 using this script"
mkdir -p "${lm_stats_dir}"; echo "${run_args} --stage 6 \"\$@\"; exit \$?" > "${lm_stats_dir}/run.sh"; chmod +x "${lm_stats_dir}/run.sh"
# 3. Submit jobs
log "LM collect-stats started... log: '${_logdir}/stats.*.log'"
# NOTE: --*_shape_file doesn't require length information if --batch_type=unsorted,
# but it's used only for deciding the sample ids.
# shellcheck disable=SC2046,SC2086
${train_cmd} JOB=1:"${_nj}" "${_logdir}"/stats.JOB.log \
${python} -m espnet2.bin.lm_train \
--collect_stats true \
--use_preprocessor true \
--bpemodel "${bpemodel}" \
--token_type "${lm_token_type}"\
--token_list "${lm_token_list}" \
--non_linguistic_symbols "${nlsyms_txt}" \
--cleaner "${cleaner}" \
--g2p "${g2p}" \
--train_data_path_and_name_and_type "${data_feats}/lm_train.txt,text,text" \
--valid_data_path_and_name_and_type "${lm_dev_text},text,text" \
--train_shape_file "${_logdir}/train.JOB.scp" \
--valid_shape_file "${_logdir}/dev.JOB.scp" \
--output_dir "${_logdir}/stats.JOB" \
${_opts} ${lm_args} || { cat $(grep -l -i error "${_logdir}"/stats.*.log) ; exit 1; }
# 4. Aggregate shape files
_opts=
for i in $(seq "${_nj}"); do
_opts+="--input_dir ${_logdir}/stats.${i} "
done
# shellcheck disable=SC2086
${python} -m espnet2.bin.aggregate_stats_dirs ${_opts} --output_dir "${lm_stats_dir}"
# Append the num-tokens at the last dimensions. This is used for batch-bins count
<"${lm_stats_dir}/train/text_shape" \
awk -v N="$(<${lm_token_list} wc -l)" '{ print $0 "," N }' \
>"${lm_stats_dir}/train/text_shape.${lm_token_type}"
<"${lm_stats_dir}/valid/text_shape" \
awk -v N="$(<${lm_token_list} wc -l)" '{ print $0 "," N }' \
>"${lm_stats_dir}/valid/text_shape.${lm_token_type}"
fi
if [ ${stage} -le 7 ] && [ ${stop_stage} -ge 7 ] && ! [[ " ${skip_stages} " =~ [[:space:]]7[[:space:]] ]]; then
log "Stage 7: LM Training: train_set=${data_feats}/lm_train.txt, dev_set=${lm_dev_text}"
_opts=
if [ -n "${lm_config}" ]; then
# To generate the config file: e.g.
# % python3 -m espnet2.bin.lm_train --print_config --optim adam
_opts+="--config ${lm_config} "
fi
if [ "${num_splits_lm}" -gt 1 ]; then
# If you met a memory error when parsing text files, this option may help you.
# The corpus is split into subsets and each subset is used for training one by one in order,
# so the memory footprint can be limited to the memory required for each dataset.
_split_dir="${lm_stats_dir}/splits${num_splits_lm}"
if [ ! -f "${_split_dir}/.done" ]; then
rm -f "${_split_dir}/.done"
${python} -m espnet2.bin.split_scps \
--scps "${data_feats}/lm_train.txt" "${lm_stats_dir}/train/text_shape.${lm_token_type}" \
--num_splits "${num_splits_lm}" \
--output_dir "${_split_dir}"
touch "${_split_dir}/.done"
else
log "${_split_dir}/.done exists. Spliting is skipped"
fi
_opts+="--train_data_path_and_name_and_type ${_split_dir}/lm_train.txt,text,text "
_opts+="--train_shape_file ${_split_dir}/text_shape.${lm_token_type} "
_opts+="--multiple_iterator true "
else
_opts+="--train_data_path_and_name_and_type ${data_feats}/lm_train.txt,text,text "
_opts+="--train_shape_file ${lm_stats_dir}/train/text_shape.${lm_token_type} "
fi
# NOTE(kamo): --fold_length is used only if --batch_type=folded and it's ignored in the other case
log "Generate '${lm_exp}/run.sh'. You can resume the process from stage 7 using this script"
mkdir -p "${lm_exp}"; echo "${run_args} --stage 7 \"\$@\"; exit \$?" > "${lm_exp}/run.sh"; chmod +x "${lm_exp}/run.sh"
log "LM training started... log: '${lm_exp}/train.log'"
if echo "${cuda_cmd}" | grep -e queue.pl -e queue-freegpu.pl &> /dev/null; then
# SGE can't include "/" in a job name
jobname="$(basename ${lm_exp})"
else
jobname="${lm_exp}/train.log"
fi
# shellcheck disable=SC2086
${python} -m espnet2.bin.launch \
--cmd "${cuda_cmd} --name ${jobname}" \
--log "${lm_exp}"/train.log \
--ngpu "${ngpu}" \
--num_nodes "${num_nodes}" \
--init_file_prefix "${lm_exp}"/.dist_init_ \
--multiprocessing_distributed true -- \
${python} -m espnet2.bin.lm_train \
--ngpu "${ngpu}" \
--use_preprocessor true \
--bpemodel "${bpemodel}" \
--token_type "${lm_token_type}"\
--token_list "${lm_token_list}" \
--non_linguistic_symbols "${nlsyms_txt}" \
--cleaner "${cleaner}" \
--g2p "${g2p}" \
--valid_data_path_and_name_and_type "${lm_dev_text},text,text" \
--valid_shape_file "${lm_stats_dir}/valid/text_shape.${lm_token_type}" \
--fold_length "${lm_fold_length}" \
--resume true \
--output_dir "${lm_exp}" \
${_opts} ${lm_args}
fi
if [ ${stage} -le 8 ] && [ ${stop_stage} -ge 8 ] && ! [[ " ${skip_stages} " =~ [[:space:]]8[[:space:]] ]]; then
log "Stage 8: Calc perplexity: ${lm_test_text}"
_opts=
# TODO(kamo): Parallelize?
log "Perplexity calculation started... log: '${lm_exp}/perplexity_test/lm_calc_perplexity.log'"
# shellcheck disable=SC2086
${cuda_cmd} --gpu "${ngpu}" "${lm_exp}"/perplexity_test/lm_calc_perplexity.log \
${python} -m espnet2.bin.lm_calc_perplexity \
--ngpu "${ngpu}" \
--data_path_and_name_and_type "${lm_test_text},text,text" \
--train_config "${lm_exp}"/config.yaml \
--model_file "${lm_exp}/${inference_lm}" \
--output_dir "${lm_exp}/perplexity_test" \
${_opts}
log "PPL: ${lm_test_text}: $(cat ${lm_exp}/perplexity_test/ppl)"
fi
if [ ${stage} -le 9 ] && [ ${stop_stage} -ge 9 ] && ! [[ " ${skip_stages} " =~ [[:space:]]9[[:space:]] ]]; then
log "Stage 9: Ngram Training: train_set=${data_feats}/lm_train.txt"
mkdir -p ${ngram_exp}
cut -f 2- -d " " ${data_feats}/lm_train.txt | lmplz -S "20%" --discount_fallback -o ${ngram_num} - >${ngram_exp}/${ngram_num}gram.arpa
build_binary -s ${ngram_exp}/${ngram_num}gram.arpa ${ngram_exp}/${ngram_num}gram.bin
fi
if [ ${stage} -le 10 ] && [ ${stop_stage} -ge 10 ] && ! [[ " ${skip_stages} " =~ [[:space:]]10[[:space:]] ]]; then
_s2t_train_dir="${data_feats}/${train_set}"
_s2t_valid_dir="${data_feats}/${valid_set}"
log "Stage 10: S2T collect stats: train_set=${_s2t_train_dir}, valid_set=${_s2t_valid_dir}"
_opts=
if [ -n "${s2t_config}" ]; then
# To generate the config file: e.g.
# % python3 -m espnet2.bin.s2t_train --print_config --optim adam
_opts+="--config ${s2t_config} "
fi
_feats_type="$(<${_s2t_train_dir}/feats_type)"
_audio_format="$(cat ${_s2t_train_dir}/audio_format 2>/dev/null || echo ${audio_format})"
if [ "${_feats_type}" = raw ]; then
_scp=wav.scp
if [[ "${_audio_format}" == *ark* ]]; then
_type=kaldi_ark
else
# "sound" supports "wav", "flac", etc.
_type=sound
fi
_opts+="--frontend_conf fs=${fs} "
else
_scp=feats.scp
_type=kaldi_ark
_input_size="$(<${_s2t_train_dir}/feats_dim)"
_opts+="--input_size=${_input_size} "
fi
# 1. Split the key file
_logdir="${s2t_stats_dir}/logdir"
mkdir -p "${_logdir}"
# Get the minimum number among ${nj} and the number lines of input files
_nj=$(min "${nj}" "$(<${_s2t_train_dir}/${_scp} wc -l)" "$(<${_s2t_valid_dir}/${_scp} wc -l)")
key_file="${_s2t_train_dir}/${_scp}"
split_scps=""
for n in $(seq "${_nj}"); do
split_scps+=" ${_logdir}/train.${n}.scp"
done
# shellcheck disable=SC2086
utils/split_scp.pl "${key_file}" ${split_scps}
key_file="${_s2t_valid_dir}/${_scp}"
split_scps=""
for n in $(seq "${_nj}"); do
split_scps+=" ${_logdir}/valid.${n}.scp"
done
# shellcheck disable=SC2086
utils/split_scp.pl "${key_file}" ${split_scps}
# 2. Generate run.sh
log "Generate '${s2t_stats_dir}/run.sh'. You can resume the process from stage 10 using this script"
mkdir -p "${s2t_stats_dir}"; echo "${run_args} --stage 10 \"\$@\"; exit \$?" > "${s2t_stats_dir}/run.sh"; chmod +x "${s2t_stats_dir}/run.sh"
# 3. Submit jobs
log "S2T collect-stats started... log: '${_logdir}/stats.*.log'"
# NOTE: --*_shape_file doesn't require length information if --batch_type=unsorted,
# but it's used only for deciding the sample ids.
_opts+="--train_data_path_and_name_and_type ${_s2t_train_dir}/${_scp},speech,${_type} "
_opts+="--valid_data_path_and_name_and_type ${_s2t_valid_dir}/${_scp},speech,${_type} "
# shellcheck disable=SC2068
for extra_txt in ${utt_extra_files}; do
_opts+="--train_data_path_and_name_and_type ${_s2t_train_dir}/${extra_txt},${extra_txt//./_},text "
_opts+="--valid_data_path_and_name_and_type ${_s2t_valid_dir}/${extra_txt},${extra_txt//./_},text "
done
for i in ${!ref_text_files[@]}; do
_opts+="--train_data_path_and_name_and_type ${_s2t_train_dir}/${ref_text_files[$i]},${ref_text_names[$i]},text "
_opts+="--valid_data_path_and_name_and_type ${_s2t_valid_dir}/${ref_text_files[$i]},${ref_text_names[$i]},text "
done
# shellcheck disable=SC2046,SC2086
${train_cmd} JOB=1:"${_nj}" "${_logdir}"/stats.JOB.log \
${python} -m espnet2.bin.s2t_train \
--collect_stats true \
--use_preprocessor true \
--bpemodel "${bpemodel}" \
--token_type "${token_type}" \
--token_list "${token_list}" \
--non_linguistic_symbols "${nlsyms_txt}" \
--cleaner "${cleaner}" \
--g2p "${g2p}" \
--train_shape_file "${_logdir}/train.JOB.scp" \
--valid_shape_file "${_logdir}/valid.JOB.scp" \
--output_dir "${_logdir}/stats.JOB" \
${_opts} ${s2t_args} || { cat $(grep -l -i error "${_logdir}"/stats.*.log) ; exit 1; }
# 4. Aggregate shape files
_opts=
for i in $(seq "${_nj}"); do
_opts+="--input_dir ${_logdir}/stats.${i} "
done
if [ "${feats_normalize}" != global_mvn ]; then
# Skip summerizaing stats if not using global MVN
_opts+="--skip_sum_stats"
fi
# shellcheck disable=SC2086
${python} -m espnet2.bin.aggregate_stats_dirs ${_opts} --output_dir "${s2t_stats_dir}"
# Append the num-tokens at the last dimensions. This is used for batch-bins count
# shellcheck disable=SC2068
for extra_txt in ${utt_extra_files}; do
_extra_txt=${extra_txt//./_}
<"${s2t_stats_dir}/train/${_extra_txt}_shape" \
awk -v N="$(<${token_list} wc -l)" '{ print $0 "," N }' \
>"${s2t_stats_dir}/train/${_extra_txt}_shape.${token_type}"
<"${s2t_stats_dir}/valid/${_extra_txt}_shape" \
awk -v N="$(<${token_list} wc -l)" '{ print $0 "," N }' \
>"${s2t_stats_dir}/valid/${_extra_txt}_shape.${token_type}"
done
for ref_txt in ${ref_text_names[@]}; do
<"${s2t_stats_dir}/train/${ref_txt}_shape" \
awk -v N="$(<${token_list} wc -l)" '{ print $0 "," N }' \
>"${s2t_stats_dir}/train/${ref_txt}_shape.${token_type}"
<"${s2t_stats_dir}/valid/${ref_txt}_shape" \
awk -v N="$(<${token_list} wc -l)" '{ print $0 "," N }' \
>"${s2t_stats_dir}/valid/${ref_txt}_shape.${token_type}"
done
fi
if [ ${stage} -le 11 ] && [ ${stop_stage} -ge 11 ] && ! [[ " ${skip_stages} " =~ [[:space:]]11[[:space:]] ]]; then
_s2t_train_dir="${data_feats}/${train_set}"
_s2t_valid_dir="${data_feats}/${valid_set}"
log "Stage 11: S2T Training: train_set=${_s2t_train_dir}, valid_set=${_s2t_valid_dir}"
_opts=
if [ -n "${s2t_config}" ]; then
# To generate the config file: e.g.
# % python3 -m espnet2.bin.s2t_train --print_config --optim adam
_opts+="--config ${s2t_config} "
fi
_feats_type="$(<${_s2t_train_dir}/feats_type)"
_audio_format="$(cat ${_s2t_train_dir}/audio_format 2>/dev/null || echo ${audio_format})"
if [ "${_feats_type}" = raw ]; then
_scp=wav.scp
# "sound" supports "wav", "flac", etc.
if [[ "${_audio_format}" == *ark* ]]; then
_type=kaldi_ark
elif [[ "${_audio_format}" == *multi* ]]; then
_type=multi_columns_sound
else
_type=sound
fi
_fold_length="$((s2t_speech_fold_length * 100))"
_opts+="--frontend_conf fs=${fs} "
else
_scp=feats.scp
_type=kaldi_ark
_fold_length="${s2t_speech_fold_length}"
_input_size="$(<${_s2t_train_dir}/feats_dim)"
_opts+="--input_size=${_input_size} "
fi
if [ "${feats_normalize}" = global_mvn ]; then
# Default normalization is utterance_mvn and changes to global_mvn
_opts+="--normalize=global_mvn --normalize_conf stats_file=${s2t_stats_dir}/train/feats_stats.npz "
fi
if [ "${num_splits_s2t}" -gt 1 ]; then
# If you met a memory error when parsing text files, this option may help you.
# The corpus is split into subsets and each subset is used for training one by one in order,
# so the memory footprint can be limited to the memory required for each dataset.
_split_dir="${s2t_stats_dir}/splits${num_splits_s2t}"
_all_scps="${_s2t_train_dir}/${_scp} ${_s2t_train_dir}/text ${s2t_stats_dir}/train/speech_shape ${s2t_stats_dir}/train/text_shape.${token_type} "
for extra_txt in ${utt_extra_files}; do
_all_scps+="${_s2t_train_dir}/${extra_txt} ${s2t_stats_dir}/train/${extra_txt//./_}_shape.${token_type} "
done
if [ ! -f "${_split_dir}/.done" ]; then
rm -f "${_split_dir}/.done"
${python} -m espnet2.bin.split_scps \
--scps ${_all_scps} \
--num_splits "${num_splits_s2t}" \
--output_dir "${_split_dir}"
touch "${_split_dir}/.done"
else
log "${_split_dir}/.done exists. Spliting is skipped"
fi
_opts+="--train_data_path_and_name_and_type ${_split_dir}/${_scp},speech,${_type} "
_opts+="--train_shape_file ${_split_dir}/speech_shape "
# shellcheck disable=SC2068
for extra_txt in ${utt_extra_files}; do
_opts+="--fold_length ${s2t_text_fold_length} "
_opts+="--train_data_path_and_name_and_type ${_split_dir}/${extra_txt},${extra_txt//./_},text "
_opts+="--train_shape_file ${_split_dir}/${extra_txt//./_}_shape.${token_type} "
done
for i in ${!ref_text_names[@]}; do
_opts+="--fold_length ${s2t_text_fold_length} "
_opts+="--train_data_path_and_name_and_type ${_split_dir}/${ref_text_files[$i]},${ref_text_names[$i]},text "
_opts+="--train_shape_file ${_split_dir}/${ref_text_names[$i]}_shape.${token_type} "
done
_opts+="--multiple_iterator true "
else
_opts+="--train_data_path_and_name_and_type ${_s2t_train_dir}/${_scp},speech,${_type} "
_opts+="--train_shape_file ${s2t_stats_dir}/train/speech_shape "
# shellcheck disable=SC2068
for extra_txt in ${utt_extra_files}; do
_opts+="--fold_length ${s2t_text_fold_length} "
_opts+="--train_data_path_and_name_and_type ${_s2t_train_dir}/${extra_txt},${extra_txt//./_},text "
_opts+="--train_shape_file ${s2t_stats_dir}/train/${extra_txt//./_}_shape.${token_type} "
done
for i in ${!ref_text_names[@]}; do
_opts+="--fold_length ${s2t_text_fold_length} "
_opts+="--train_data_path_and_name_and_type ${_s2t_train_dir}/${ref_text_files[$i]},${ref_text_names[$i]},text "
_opts+="--train_shape_file ${s2t_stats_dir}/train/${ref_text_names[$i]}_shape.${token_type} "
done
fi
# shellcheck disable=SC2068
for extra_txt in ${utt_extra_files}; do
_opts+="--valid_data_path_and_name_and_type ${_s2t_valid_dir}/${extra_txt},${extra_txt//./_},text "
_opts+="--valid_shape_file ${s2t_stats_dir}/valid/${extra_txt//./_}_shape.${token_type} "
done
for i in ${!ref_text_names[@]}; do
_opts+="--valid_data_path_and_name_and_type ${_s2t_valid_dir}/${ref_text_files[$i]},${ref_text_names[$i]},text "
_opts+="--valid_shape_file ${s2t_stats_dir}/valid/${ref_text_names[$i]}_shape.${token_type} "
done
log "Generate '${s2t_exp}/run.sh'. You can resume the process from stage 11 using this script"
mkdir -p "${s2t_exp}"; echo "${run_args} --stage 11 \"\$@\"; exit \$?" > "${s2t_exp}/run.sh"; chmod +x "${s2t_exp}/run.sh"
# NOTE(kamo): --fold_length is used only if --batch_type=folded and it's ignored in the other case
log "S2T training started... log: '${s2t_exp}/train.log'"
if echo "${cuda_cmd}" | grep -e queue.pl -e queue-freegpu.pl &> /dev/null; then
# SGE can't include "/" in a job name
jobname="$(basename ${s2t_exp})"
else
jobname="${s2t_exp}/train.log"
fi
# shellcheck disable=SC2086
${python} -m espnet2.bin.launch \
--cmd "${cuda_cmd} --name ${jobname}" \
--log "${s2t_exp}"/train.log \
--ngpu "${ngpu}" \
--num_nodes "${num_nodes}" \
--init_file_prefix "${s2t_exp}"/.dist_init_ \
--multiprocessing_distributed true -- \
${python} -m espnet2.bin.${s2t_task}_train \
--use_preprocessor true \
--bpemodel "${bpemodel}" \
--token_type "${token_type}" \
--token_list "${token_list}" \
--non_linguistic_symbols "${nlsyms_txt}" \
--cleaner "${cleaner}" \
--g2p "${g2p}" \
--valid_data_path_and_name_and_type "${_s2t_valid_dir}/${_scp},speech,${_type}" \
--valid_shape_file "${s2t_stats_dir}/valid/speech_shape" \
--resume true \
--fold_length "${_fold_length}" \
--output_dir "${s2t_exp}" \
${_opts} ${s2t_args}
fi
if [ -n "${download_model}" ]; then
log "Use ${download_model} for decoding and evaluation"
s2t_exp="${expdir}/${download_model}"
mkdir -p "${s2t_exp}"
# If the model already exists, you can skip downloading
espnet_model_zoo_download --unpack true "${download_model}" > "${s2t_exp}/config.txt"
# Get the path of each file
_s2t_model_file=$(<"${s2t_exp}/config.txt" sed -e "s/.*'s2t_model_file': '\([^']*\)'.*$/\1/")
_s2t_train_config=$(<"${s2t_exp}/config.txt" sed -e "s/.*'s2t_train_config': '\([^']*\)'.*$/\1/")
# Create symbolic links
ln -sf "${_s2t_model_file}" "${s2t_exp}"
ln -sf "${_s2t_train_config}" "${s2t_exp}"
inference_s2t_model=$(basename "${_s2t_model_file}")
if [ "$(<${s2t_exp}/config.txt grep -c lm_file)" -gt 0 ]; then
_lm_file=$(<"${s2t_exp}/config.txt" sed -e "s/.*'lm_file': '\([^']*\)'.*$/\1/")
_lm_train_config=$(<"${s2t_exp}/config.txt" sed -e "s/.*'lm_train_config': '\([^']*\)'.*$/\1/")
lm_exp="${expdir}/${download_model}/lm"
mkdir -p "${lm_exp}"
ln -sf "${_lm_file}" "${lm_exp}"
ln -sf "${_lm_train_config}" "${lm_exp}"
inference_lm=$(basename "${_lm_file}")
fi
fi
if [ ${stage} -le 12 ] && [ ${stop_stage} -ge 12 ] && ! [[ " ${skip_stages} " =~ [[:space:]]12[[:space:]] ]]; then
log "Stage 12: Decoding: training_dir=${s2t_exp}"
if ${gpu_inference}; then
_cmd="${cuda_cmd}"
_ngpu=1
else
_cmd="${decode_cmd}"
_ngpu=0
fi
_opts=
if [ -n "${inference_config}" ]; then
_opts+="--config ${inference_config} "
fi
if "${use_lm}"; then
if "${use_word_lm}"; then
_opts+="--word_lm_train_config ${lm_exp}/config.yaml "
_opts+="--word_lm_file ${lm_exp}/${inference_lm} "
else
_opts+="--lm_train_config ${lm_exp}/config.yaml "
_opts+="--lm_file ${lm_exp}/${inference_lm} "
fi
fi
if "${use_ngram}"; then
_opts+="--ngram_file ${ngram_exp}/${inference_ngram}"
fi
# 2. Generate run.sh
log "Generate '${s2t_exp}/${inference_tag}/run.sh'. You can resume the process from stage 12 using this script"
mkdir -p "${s2t_exp}/${inference_tag}"; echo "${run_args} --stage 12 \"\$@\"; exit \$?" > "${s2t_exp}/${inference_tag}/run.sh"; chmod +x "${s2t_exp}/${inference_tag}/run.sh"
inference_bin_tag=""
if "${use_streaming}"; then
inference_bin_tag="_streaming"
fi
if "${eval_valid_set}"; then
_dsets="org/${valid_set} ${test_sets}"
else
_dsets="${test_sets}"
fi
for dset in ${_dsets}; do
_data="${data_feats}/${dset}"
_dir="${s2t_exp}/${inference_tag}/${dset}"
_logdir="${_dir}/logdir"
mkdir -p "${_logdir}"
_feats_type="$(<${_data}/feats_type)"
_audio_format="$(cat ${_data}/audio_format 2>/dev/null || echo ${audio_format})"
if [ "${_feats_type}" = raw ]; then
_scp=wav.scp
if [[ "${audio_format}" == *ark* ]]; then
_type=kaldi_ark
elif [[ "${_audio_format}" == *multi* ]]; then
_type=multi_columns_sound
else
_type=sound
fi
else
_scp=feats.scp
_type=kaldi_ark
fi
# 1. Split the key file
key_file=${_data}/${_scp}
split_scps=""
_nj=$(min "${inference_nj}" "$(<${key_file} wc -l)")
for n in $(seq "${_nj}"); do
split_scps+=" ${_logdir}/keys.${n}.scp"
done
# shellcheck disable=SC2086
utils/split_scp.pl "${key_file}" ${split_scps}
# 2. Submit decoding jobs
log "Decoding started... log: '${_logdir}/s2t_inference.*.log'"
rm -f "${_logdir}/*.log"
# shellcheck disable=SC2046,SC2086
${_cmd} --gpu "${_ngpu}" JOB=1:"${_nj}" "${_logdir}"/s2t_inference.JOB.log \
${python} -m espnet2.bin.${s2t_task}_inference${inference_bin_tag} \
--batch_size ${batch_size} \
--ngpu "${_ngpu}" \
--data_path_and_name_and_type "${_data}/${_scp},speech,${_type}" \
--key_file "${_logdir}"/keys.JOB.scp \
--s2t_train_config "${s2t_exp}"/config.yaml \
--s2t_model_file "${s2t_exp}"/"${inference_s2t_model}" \
--output_dir "${_logdir}"/output.JOB \
${_opts} ${inference_args} || { cat $(grep -l -i error "${_logdir}"/s2t_inference.*.log) ; exit 1; }
# 3. Concatenates the output files from each jobs
# shellcheck disable=SC2068
for ref_txt in ${ref_text_files[@]}; do
suffix=$(echo ${ref_txt} | sed 's/text//')
for f in token token_int score text text_nospecial; do
if [ -f "${_logdir}/output.1/1best_recog/${f}${suffix}" ]; then
for i in $(seq "${_nj}"); do
cat "${_logdir}/output.${i}/1best_recog/${f}${suffix}"
done | sort -k1 >"${_dir}/${f}${suffix}"
fi
done
done
done
fi
if [ ${stage} -le 13 ] && [ ${stop_stage} -ge 13 ] && ! [[ " ${skip_stages} " =~ [[:space:]]13[[:space:]] ]]; then
log "Stage 13: Scoring"
if [ "${token_type}" = phn ]; then
log "Error: Not implemented for token_type=phn"
exit 1
fi
if "${eval_valid_set}"; then
_dsets="org/${valid_set} ${test_sets}"
else
_dsets="${test_sets}"
fi
for dset in ${_dsets}; do
_data="${data_feats}/${dset}"
_dir="${s2t_exp}/${inference_tag}/${dset}"
for _tok_type in "char" "word" "bpe"; do
[ "${_tok_type}" = bpe ] && [ ! -f "${bpemodel}" ] && continue
_opts="--token_type ${_tok_type} "
if [ "${_tok_type}" = "char" ] || [ "${_tok_type}" = "word" ]; then
_type="${_tok_type:0:1}er"
_opts+="--non_linguistic_symbols ${nlsyms_txt} "
_opts+="--remove_non_linguistic_symbols true "
elif [ "${_tok_type}" = "bpe" ]; then
_type="ter"
_opts+="--bpemodel ${bpemodel} "
else
log "Error: unsupported token type ${_tok_type}"
fi
_scoredir="${_dir}/score_${_type}"
mkdir -p "${_scoredir}"
# shellcheck disable=SC2068
for ref_txt in ${ref_text_files[@]}; do
# Note(simpleoier): to get the suffix after text, e.g. "text_spk1" -> "_spk1"
suffix=$(echo ${ref_txt} | sed 's/text//')
# Tokenize text to ${_tok_type} level
paste \
<(<"${_data}/${ref_txt}" \
${python} -m espnet2.bin.tokenize_text \
-f 2- --input - --output - \
--cleaner "${cleaner}" \
${_opts} \
) \
<(<"${_data}/utt2spk" awk '{ print "(" $2 "-" $1 ")" }') \
>"${_scoredir}/ref${suffix:-${suffix}}.trn"
paste \
<(<"${_dir}/${ref_txt}_nospecial" \
${python} -m espnet2.bin.tokenize_text \
-f 2- --input - --output - \
${_opts} \
--cleaner "${hyp_cleaner}" \
) \
<(<"${_data}/utt2spk" awk '{ print "(" $2 "-" $1 ")" }') \
>"${_scoredir}/hyp${suffix:-${suffix}}.trn"
done
#sclite \
#${score_opts} \
#-r "${_scoredir}/ref.trn" trn \
#-h "${_scoredir}/hyp.trn" trn \
#-i rm -o all stdout > "${_scoredir}/result.txt"
#log "Write ${_type} result in ${_scoredir}/result.txt"
#grep -e Avg -e SPKR -m 2 "${_scoredir}/result.txt"
done
done
[ -f local/score.sh ] && local/score.sh ${local_score_opts} "${s2t_exp}"
# Show results in Markdown syntax
scripts/utils/show_asr_result.sh "${s2t_exp}" > "${s2t_exp}"/RESULTS.md
cat "${s2t_exp}"/RESULTS.md
fi
packed_model="${s2t_exp}/${s2t_exp##*/}_${inference_s2t_model%.*}.zip"
if [ ${stage} -le 14 ] && [ ${stop_stage} -ge 14 ] && ! [[ " ${skip_stages} " =~ [[:space:]]14[[:space:]] ]]; then
log "Stage 14: Pack model: ${packed_model}"
_opts=
if "${use_lm}"; then
_opts+="--lm_train_config ${lm_exp}/config.yaml "
_opts+="--lm_file ${lm_exp}/${inference_lm} "
_opts+="--option ${lm_exp}/perplexity_test/ppl "
_opts+="--option ${lm_exp}/images "
fi
if [ "${feats_normalize}" = global_mvn ]; then
_opts+="--option ${s2t_stats_dir}/train/feats_stats.npz "
fi
if [ "${token_type}" = bpe ]; then
_opts+="--option ${bpemodel} "
fi
if [ "${nlsyms_txt}" != none ]; then
_opts+="--option ${nlsyms_txt} "
fi
# shellcheck disable=SC2086
${python} -m espnet2.bin.pack s2t \
--s2t_train_config "${s2t_exp}"/config.yaml \
--s2t_model_file "${s2t_exp}"/"${inference_s2t_model}" \
${_opts} \
--option "${s2t_exp}"/RESULTS.md \
--option "${s2t_exp}"/images \
--outpath "${packed_model}"
fi
if [ ${stage} -le 15 ] && [ ${stop_stage} -ge 15 ] && ! [[ " ${skip_stages} " =~ [[:space:]]15[[:space:]] ]]; then
[ -z "${hf_repo}" ] && \
log "ERROR: You need to setup the variable hf_repo with the name of the repository located at HuggingFace, follow the following steps described here https://github.com/espnet/espnet/blob/master/CONTRIBUTING.md#132-espnet2-recipes" && \
exit 1
log "Stage 15: Upload model to HuggingFace: ${hf_repo}"
if [ ! -f "${packed_model}" ]; then
log "ERROR: ${packed_model} does not exist. Please run stage 14 first."
exit 1
fi
gitlfs=$(git lfs --version 2> /dev/null || true)
[ -z "${gitlfs}" ] && \
log "ERROR: You need to install git-lfs first" && \
exit 1
dir_repo=${expdir}/hf_${hf_repo//"/"/"_"}
[ ! -d "${dir_repo}" ] && git clone https://huggingface.co/${hf_repo} ${dir_repo}
if command -v git &> /dev/null; then
_creator_name="$(git config user.name)"
_checkout="git checkout $(git show -s --format=%H)"
else
_creator_name="$(whoami)"
_checkout=""
fi
# /some/where/espnet/egs2/foo/s2t1/ -> foo/s2t1
_task="$(pwd | rev | cut -d/ -f2 | rev)"
# foo/s2t1 -> foo
_corpus="${_task%/*}"
_model_name="${_creator_name}/${_corpus}_$(basename ${packed_model} .zip)"
# copy files in ${dir_repo}
unzip -o ${packed_model} -d ${dir_repo}
# Generate description file
# shellcheck disable=SC2034
hf_task=automatic-speech-recognition
# shellcheck disable=SC2034
espnet_task=S2T
# shellcheck disable=SC2034
task_exp=${s2t_exp}
eval "echo \"$(cat scripts/utils/TEMPLATE_HF_Readme.md)\"" > "${dir_repo}"/README.md
this_folder=${PWD}
cd ${dir_repo}
if [ -n "$(git status --porcelain)" ]; then
git add .
git commit -m "Update model"
fi
git push
cd ${this_folder}
fi
log "Successfully finished. [elapsed=${SECONDS}s]"
|