Blame view
egs/hub4_spanish/s5/local/lexicon/make_unicode_lexicon.py
26.4 KB
8dcb6dfcb
first commit
|
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 |
#!/usr/bin/env python
# -*- coding: utf-8 -*-
# Copyright 2016 Johns Hopkins University (Author: Matthew Wiesner)
# Apache 2.0
# ============ Make unicode-based graphemic lexicon =============
#
# This script takes a list of either words or words and corresponding
# morphemes and returns a graphemic lexicon in the "standard" kaldi format,
# i.e. a single word with its correpsonding pronunciation per line; multiple
# pronunciations of a word are listed on separate lines.
#
# Example:
# word w o r d
# word w1 o r d1
# anotherword a n o t h e r w o r d
#
# It also creates a mapping file describing how each grapheme is transformed
# into graphemic acoustic units. It has the following form:
#
# Example:
# በ b a
# â a_combiningcircumflex
# ü ucombiningdiaeresis
# b b
# c c
#
# When the script is called with the option (--apply-map mapfile) the map
# provided in "mapfile" is used to expand the words in the provided wordlist
# into graphemic-acoustic units and the resulting lexicon is output along with
# the input mapfile used.
#
#
# When the script is called with the option (-V LOGDIR) this is interpreted as
# the directory into which two log files are stored. These log files contain
# information about grapheme frequencies in the vocabulary as well as a table
# that stores the information about how to map graphemes into the graphemic
# acoustic units used in the lexicon -- the atoms used to describe the
# pronunciations of words in the lexicon. The table also stores information
# about the different fields of the standard unicode description of a grapheme.
# This description contains information about the case (CAPITAL , SMALL),
# writing system (LATIN, ARABIC, etc.), type of grapheme (LETTER, SYLLABLE,
# VOWEL SIGN, etc.), the name of the grapheme (a, b, aleph, lambda, etc.), and
# a description of any diacritics modifying the "base" grapheme. Empty fields
# are denoted by ''. These modifying diacritics are just combining characters
# in the unicode NFKD form of each grapheme's unicode character description.
# The table looks as follows
# --------------------------------------------------------------------
# CASE CHAR_TYPE LANGUAGE MAP0 MAP1 NAME SYMBOL TAG TAG0 TAG1
# SMALL LETTER LATIN u '' U u '' '' ''
# CAPITAL LETTER LATIN u '' U U '' '' ''
# SMALL LETTER LATIN ucombiningacuteaccent '' U ú WITH ACUTE COMBINING ACUTE ACCENT ''
# SMALL LETTER LATIN u_combiningdiaeresis '' U ü WITH DIAERESIS COMBINING DIAERESIS ''
# '' SYLLABLE ETHIOPIC b a BA በ '' '' ''
# SMALL LETTER LATIN a_combiningbreve_combiningacuteaccent '' A ắ WITH BREVE AND ACUTE COMBINING BREVE COMBINING ACUTE ACCENT
# ---------------------------------------------------------------------
# Below is a summary of the fields and their meanings:
# -----------------------------------------------------
# CASE: Case of the grapheme (CAPITAL, SMALL)
# CHAR_TYPE: The type of grapheme (LETTER, SYLLABLE, VOWEL SIGN, etc.).
# This field sometimes determines the way the character is mapped;
# a syllable for instance would result in a one-to-many mapping.
#
# LANGUAGE: The script from which the grapheme originated.
# Examples are LATIN, ETHIOPIC, or KATAKANA.
# MAP0: The first acoustic unit to which the grapheme maps.
# MAP1, MAP2, ...: The subsequent acoustic units to which a given grapheme maps.
# NAME: The name of the base grapheme in the unicode description.
# SYMBOL: The actual grapheme.
# TAG: The unicode description of any diacritics attached to the base
# grapheme in the unicode description.
# TAG0: The name of the first combining character in the NFKD form for the
# unicode character.
# TAG1, TAG2, ...: The name of subsequent combining characters in the NFKD
# form of the grapheme.
# -------------------------------------------------------------
#
# The mapping is assumed (for now) to be one-to-one or one-to-many. If a single
# grapheme gets mapped to multiple acoustic units, the units are stored in the
# the fields MAP0, then MAP1, MAP2, etc., depending on the number of acoustic
# units generated by the grapheme. This normally occurs for syllabaries or
# abugidas where each grapheme represents a syllable, and hence more than a
# single phoneme.
# Similarly a base grapheme may have more than one diacritic. The name of each
# diactritic, represented as a combining character in the NFKD baseform, is
# stored in order in the fields TAG0, TAG1, etc.. A field is present in the
# table if any grapheme occurring in the vocabulary has said field.
# The other log file shows each grapheme in the vocabulary, it's relative
# frequency, and the threshold frequency above which each grapheme-diactritic
# combination is treated as a separate unit. We include the use of tags on
# acoustic units for sufficiently rare grapheme-diacritic combinations. They
# are represented by the name of the base grapheme followed by an underscore
# and the name of the combining character. The threshold for deciding which
# grapheme-diacritic combinations result in tagged units rather than a
# completely distinct unit is an option to the script (-T). Using -T 1.0,
# results in all combining characters being treated as tags. Using -T 0.0 means
# that no acoustic units are tagged and each grapheme-combining character
# combination results in a distinct acoustic unit.
# ===============================================================
# Import Statements
from __future__ import print_function
from __future__ import division
import codecs
import argparse
import unicodedata
import os
import re
import sys
import numpy as np
def main():
args = parse_input()
baseforms = get_word_list(args.word2baseform)
if args.apply_map:
grapheme_map = {}
with codecs.open(args.apply_map, "r", encoding="utf-8") as f:
for line in f:
try:
line_vals = line.strip('
').split(' ', 1)
grapheme_map[line_vals[0]] = line_vals[1]
except IndexError:
grapheme_map[line_vals[0]] = ""
encoded_transcription = apply_map(grapheme_map, baseforms)
else:
unicode_transcription = baseform2unicode(baseforms)
encoded_transcription, table, grapheme_map = encode(unicode_transcription,
args.tag_percentage,
log=args.verbose)
if args.verbose:
if not os.path.exists(args.verbose):
os.makedirs(args.verbose)
write_table(table, os.path.join(args.verbose, "grapheme_table.txt"))
# Extract nonspeech lexicon (e.g. <laugh>, <silence>, <cough>)
try:
silence_lexicon = {}
with codecs.open(args.silence_lexicon, "r", "utf-8") as f:
for line in f:
line_vals = line.strip().split(None, 1)
silence_lexicon[line_vals[0]] = line_vals[1]
except (IOError, TypeError):
pass
# Extract dictionary of extraspeech pronunciations (normally <hes>)
try:
extra_lexicon = {}
with codecs.open(args.extra_lexicon, "r", "utf-8") as f:
for line in f:
line_vals = line.strip().split(None, 1)
extra_lexicon[line_vals[0]] = line_vals[1]
except (IOError, TypeError):
pass
write_map(grapheme_map, args.map_out)
write_lexicon(baseforms, encoded_transcription, args.lexicon_out,
sil_lex=silence_lexicon, extra_lex=extra_lexicon)
def parse_input():
'''
Parse commandline input.
'''
if len(sys.argv[1:]) == 0:
print("Usage: ./make_unicode_lexicon.py [opts] lex_in lex_out")
sys.exit(1)
parser = argparse.ArgumentParser()
parser.add_argument("word2baseform", help="File with word list optionally"
" paired with a baseform. 1 word per line with the "
"baseform separated by a tab")
parser.add_argument("lexicon_out", help="Path of output graphemc lexicon")
parser.add_argument("map_out", help="Path of output "
"grapheme-to-graphemic-acoustic units map")
parser.add_argument("-T", "--tag-percentage", help="Percentage of least"
" frequently occurring graphemes to be tagged",
type=float, action="store", default=0.1)
parser.add_argument("--silence-lexicon", help="File with silence words "
"and pronunciations", action="store", default=None)
parser.add_argument("--extra-lexicon", help="File with extra speech words "
"and pronunciations", action="store", default=None)
parser.add_argument("-V", "--verbose", help="Directory for storing useful "
"log files", action="store", default=None)
parser.add_argument("--apply-map", help="Map to apply to wordlist",
action="store", default=None)
args = parser.parse_args()
return args
def _read_word_list_line(line):
try:
word2baseform = line.strip().split(None, 1)
return (word2baseform[0], word2baseform[1])
except IndexError:
return (word2baseform[0], word2baseform[0])
def get_word_list(input_file):
'''
Read from input file the words and potential baseforms.
Arguments: input_file -- path to the input word list optionally with
baseforms (1 per line word baseform).
Output:
words -- list of tuples (word, baseform)
'''
with codecs.open(input_file, "r", "utf-8") as f:
words = []
for line in f:
w = _read_word_list_line(line)
words.append(w)
return words
def baseform2unicode(baseforms):
'''
Convert each baseform in the list, baseforms, to a parsed unicode
description stored as a list of lists of dictionaries.
unicode_transcription = [
[{'NAME':'word1_grapheme1','FIELD1':'FIELD1_VAL',...},
{'NAME':'word1_grapheme2','FIELD1':'FIELD1_VAL',...},...],
[{'NAME':'word2_grapheme1,'FIELD1:'FIELD1_VAL',...},
{},...]
,...,[]]
Arguments:
baseforms -- List of tuples (word, baseform)
e.g. baseforms = get_word_list()
Output:
unicode_transcription -- See above description
'''
# Regular expression for parsing unicode descriptions
pattern = re.compile(
r"(?P<LANGUAGE>[^\s]+)\s"
r"(?P<CASE>SMALL\s|CAPITAL\s)?(?P<CHAR_TYPE>"
"(?:SUBJOINED )?LETTER |(?:INDEPENDENT VOWEL )"
r"|(?:VOWEL SIGN )|VOWEL |SIGN "
r"|CHARACTER |JONGSEONG |CHOSEONG |SYMBOL |MARK |DIGIT "
r"|SEMIVOWEL |TONE |SYLLABLE |LIGATURE |KATAKANA )"
r"(?P<NAME>((?!WITH).)+)"
r"(?P<TAG>WITH .+)?"
)
# For each graphemic baseform generate a parsed unicode description
unicode_transcription = []
for w, bf in baseforms:
# Initialize empty list of words
baseform_transcription = []
# For each grapheme parse the unicode description
for graph in bf:
unicode_desc = unicodedata.name(graph)
# Use the canonical unicode decomposition
tags = unicodedata.normalize('NFD', graph)
match_obj = pattern.match(unicode_desc)
# Grapheme's unicode description is non-standard
if(not match_obj):
# Underscore, dash, hastag have special meaning
if(graph in ("_", "-", "#")):
graph_dict = {
'CHAR_TYPE': 'LINK',
'SYMBOL': graph,
'NAME': graph
}
# The grapheme is whitespace
elif(unicode_desc in ("ZERO WIDTH SPACE",
"ZERO WIDTH NON-JOINER",
"ZERO WIDTH JOINER",
"SPACE")):
# Ignore whitespace
continue
else:
graph_dict = {'SYMBOL': graph, 'NAME': 'NOT_FOUND'}
# Grapheme's unicode description is standard
else:
graph_dict = match_obj.groupdict()
graph_dict["SYMBOL"] = graph
# Add tags to dictionary (The first element of tags is actually
# the base grapheme, so we only check all tags after the first.
if(len(tags) > 1):
for i, t in enumerate(tags[1:]):
graph_dict["TAG" + str(i)] = unicodedata.name(t)
# Add grapheme unicode description dictionary to baseform list
baseform_transcription.append(graph_dict)
# Add baseform transcription to unicode transcription list
unicode_transcription.append(baseform_transcription)
return unicode_transcription
def encode(unicode_transcription, tag_percentage, log=False):
'''
Arguments:
unicode_transcription -- a list of words whose graphemes are
respresented as a list of dictionaries whose
fields contain information about parsed
unicode descriptions.
tag_percentage -- percent of least frequent graphemes to tag
log -- optional printing
Outputs:
encoded_transcription -- baseforms mapped to the graphemeic
acoustic units
'''
# Constants
VOWELS = "AEIOU"
SKIP = "/()"
table = []
graphemes = []
encoded_transcription = []
# Accumulate grapheme statistics over corpus at some point. For now just
# use the lexicon word list. For estimating grapheme frequency this is
# probably sufficient since we have many words each with many
# graphemes. We do unfortunately have to assume that case does not matter.
# We do not count dashes, underscores, parentheses, etc. . Just letters.
graph_list = []
for w in unicode_transcription:
for graph in w:
if graph["SYMBOL"] not in "()\/,-_#.":
graph_list.append(graph["SYMBOL"].lower())
graph2int = {v: k for k, v in enumerate(set(graph_list))}
int2graph = {v: k for k, v in graph2int.items()}
graph_list_int = [graph2int[g] for g in graph_list]
bin_edges = list(range(0, len(int2graph.keys()) + 1))
graph_counts = np.histogram(graph_list_int, bins=bin_edges)[0]/ float(len(graph_list_int))
# Set count threshold to frequency that tags the bottom 10% of graphemes
bottom_idx = int(np.floor(tag_percentage * len(graph_counts)))
count_thresh = sorted(graph_counts)[bottom_idx]
graph_counts_dict = {}
for i, count in enumerate(graph_counts):
graph_counts_dict[int2graph[i]] = count
graph_counts = graph_counts_dict
# Print grapheme counts to histogram
if log:
graph_counts_sorted = sorted(graph_counts, reverse=True,
key=graph_counts.get)
if not os.path.exists(log):
os.makedirs(log)
with codecs.open(os.path.join(log, "grapheme_histogram.txt"), "w", "utf-8") as fp:
fp.write("Graphemes (Count Threshold = %.6f) (Tag Percentage "
"= %.2f)
" % (count_thresh, tag_percentage))
for g in graph_counts_sorted:
weight = ("-" * int(np.ceil(500.0 * graph_counts[g])) +
" %.6f
" % graph_counts[g])
fp.write("%s -" % (g) + weight)
# Find a new baseform for each word
for w in unicode_transcription:
word_transcription = ""
# Find a "pronunciation" for each grapheme in the word
for graph in w:
# Case 1: Check that the grapheme has a unicode description type
# ---------------------------------------------------------------
if("CHAR_TYPE" not in [k.strip() for k in graph.keys()]):
if(graph["SYMBOL"] == "."):
try:
graph["MAP0"] = "\t"
if word_transcription[-1] == " ":
word_transcription = word_transcription[:-1] + "\t"
except IndexError:
print("Word starting with . detected")
graph["MAP0"] = "."
word_transcription = ". "
elif(graph["SYMBOL"] not in SKIP):
graph["MAP0"] = graph["SYMBOL"].lower()
word_transcription += graph["MAP0"] + " "
# Case 2: Standard Grapheme
# ---------------------------------------------------------------
elif(graph["CHAR_TYPE"].strip() in
("LETTER", "VOWEL", "VOWEL SIGN", "SIGN")):
# Backoff diacritics
base_grapheme = graph["NAME"].strip().replace(" ", "-").lower()
graph["MAP0"] = _backoff_diacritics(graph["SYMBOL"].lower(),
base_grapheme,
graph_counts,
count_thresh)
# Add final space
word_transcription += graph["MAP0"] + " "
# Case 3: Syllable (Assume consonant vowel pattern)
# At some point we will make it (cvc), but for now
# this is basically just here for Amharic
# ----------------------------------------------------------------
elif(graph["CHAR_TYPE"].strip() == "SYLLABLE"):
# Multi-word description
if(len(graph["NAME"].strip().split(' ')) > 1):
g_name = graph["NAME"].strip().replace(" ", "-").lower()
graph["MAP0"] = g_name + "\t"
word_transcription += graph["MAP0"]
# Consonant Vowel Pattern
else:
cv_pattern = (r"(?P<CONSONANT>[^%s]*)(?P<VOWEL>[%s]+)" %
(VOWELS, VOWELS))
parsed_graph = re.match(cv_pattern, graph["NAME"])
if(not parsed_graph):
sys.exit("Syllable did not obey"
"consonant-vowel pattern.")
graph_dict = parsed_graph.groupdict()
# Get consonant if it exists
if("CONSONANT" in graph_dict.keys() and
graph_dict["CONSONANT"]):
graph["MAP0"] = graph_dict["CONSONANT"].lower()
word_transcription += graph["MAP0"] + " "
# Get vowel if it exists
if("VOWEL" in graph_dict.keys() and graph_dict["VOWEL"]):
graph["MAP1"] = graph_dict["VOWEL"].lower() + "\t"
word_transcription += graph["MAP1"]
# Case 4: Commonly occurring symbols
# ----------------------------------------------------------------
elif(graph["CHAR_TYPE"].strip() == "LINK"):
# Add tab for underscores (kaldi lexicon format)
if(graph["SYMBOL"] in ("_", "#")):
graph["MAP0"] = "\t"
if(len(word_transcription) >= 3 and
word_transcription[-2] == "\t"):
word_transcription = word_transcription[:-3] + "\t"
elif(len(word_transcription) >= 1):
word_transcription += "\t"
else:
sys.exit("Unknown rule for initial underscore")
elif(graph["SYMBOL"] == "-"):
graph["MAP0"] = "\t"
else:
sys.exit("Unknown linking symbol found.")
sys.exit(1)
# Update table of observed graphemes
if(graph["SYMBOL"] not in graphemes):
table.append(graph)
graphemes.append(graph["SYMBOL"])
# Append the newly transcribed word
encoded_transcription.append(word_transcription.strip())
# Create grapheme to graphemic-acoustic-unit map
grapheme_map = {}
for g_dict in table:
g_map = ""
map_number = 0
for g_field, g_val in sorted(g_dict.items()):
if(g_field == ("MAP" + str(map_number))):
g_map = g_map + g_val + " "
map_number = map_number + 1
grapheme_map[g_dict["SYMBOL"]] = g_map.strip(' ')
return encoded_transcription, table, grapheme_map
def _backoff_diacritics(grapheme, base_grapheme, graph_counts, count_thresh):
'''
Add diacritics as tags if the grapheme with diacritics occurs
infrequently. The grapheme built by successively peeling away
diacritics until a frequent grapheme in the lexicon is discovered.
This grapheme is then considered a distinct unit and all peeled off
diacritics are added as kaldi style tags
Arguments:
grapheme -- the raw grapheme to be processed
base_grapheme -- the grapheme with no combining marks
(see unicode normalization NFD for more details)
graph_counts -- A dictionary of all seen graphemes as keys with
counts as values
count_thresh -- The frequency threshold below which diacritics
should be peeled away
'''
# Initialize variables before loop
new_grapheme = grapheme
removed = []
parts = unicodedata.normalize("NFD", new_grapheme)
# Find a backed-off (in terms of number of diacritics) grapheme with count
# above the frequency threshold (count_thresh)
while(len(parts) > 1 and
(graph_counts[new_grapheme] <= count_thresh)):
new_grapheme = unicodedata.normalize("NFC", parts[0:-1])
tag = unicodedata.name(parts[-1]).strip().replace(" ", "").lower()
removed.append(tag)
parts = unicodedata.normalize("NFD", new_grapheme)
# Collect all diactritics that will not be added as tags
split_tags = []
for p in parts[1:]:
split_tag = unicodedata.name(p).strip().replace(" ", "").lower()
split_tags.append(split_tag)
# Append non-tag diacritics to the base grapheme
base_grapheme = "".join([base_grapheme] + split_tags)
# Return the tagged grapheme
return "_".join([base_grapheme] + removed)
def apply_map(grapheme_map, baseforms):
'''
Apply the grapheme_map to the baseforms
Arguments:
grapheme_map -- dictionary storing mapping from grapheme to
graphemic-acoustic units
baseforms -- the words to which we want to apply the mappings
Outputs:
encoded_transcription -- See encode (function). It's the exact same
format.
'''
encoded_transcription = []
for w, bf in baseforms:
word_transcription = ""
for graph in bf:
try:
if grapheme_map[graph][-1] == "\t":
word_transcription += grapheme_map[graph]
else:
word_transcription += grapheme_map[graph] + " "
except KeyError:
pass
encoded_transcription.append(word_transcription.strip())
return encoded_transcription
def write_table(table, outfile):
'''
Creates table of graphemes and fields of each grapheme's corresponding
unicode description.
Arguments:
table -- table to write
outfile -- name of the output lexicon debug file
'''
# Create output table name
#outfile = os.path.splitext(outfile)[0]
# Sort keys for convenience
table_sorted = sorted(table, key=lambda k: k["NAME"])
# Start writing to output
with codecs.open(outfile, "w", "utf-8") as fo:
# Get header names
header_names = sorted(set().union(*[d.keys() for d in table]))
# Write headers
for h in header_names[:-1]:
fo.write("%s\t" % h)
fo.write("%s
" % header_names[-1])
# Write values if present
for t in table_sorted:
for h in header_names[:-1]:
if(h in t.keys() and t[h]):
fo.write("%s\t" % t[h])
else:
fo.write("''\t")
if(header_names[-1] in t.keys() and t[header_names[-1]]):
fo.write("%s
" % t[header_names[-1]])
else:
fo.write("''
")
def write_map(grapheme_map, mapfile):
'''
Write out a file with the mapping from graphemes to
graphemic-acoustic units. The format is one grapheme per line
followed by a space and then the graphemic acoustic units to which
the grapheme was mapped. Compatible with utils/apply_map.pl
Arguments:
grapheme_map -- dictionary mapping graphemes to graphemic-acoustic
units as output by encode()
mapfile -- the path to whch the grapheme map will be written
'''
with codecs.open(mapfile, 'w', encoding='utf-8') as f:
for g, g_map in grapheme_map.items():
print(g, g_map, file=f)
def write_lexicon(baseforms, encoded_transcription, outfile, sil_lex=None,
extra_lex=None):
'''
Write out the encoded transcription of words
Arguments:
baseforms -- list of words from a word list
encoded_transcription -- input encoded lexicon
outfile -- output lexicon
'''
# Write Lexicon File
with codecs.open(outfile, "w", "utf-8") as f:
# First write the non-speech words
try:
for w in sil_lex.keys():
f.write("%s\t%s
" % (w, sil_lex[w]))
except AttributeError:
pass
# Then write extra-speech words
try:
for w in extra_lex.keys():
f.write("%s\t%s
" % (w, extra_lex[w]))
except AttributeError:
pass
# Then write the rest of the words
for idx, w in enumerate(baseforms):
# This is really just for BABEL in case <hes> is written as a word
if(w[0].lower() == "<hes>"):
f.write("%s\t<hes>
" % (str(w[0])))
else:
f.write("%s\t%s
" % (str(w[0]),
encoded_transcription[idx]))
if __name__ == "__main__":
main()
|