Compare commits

..

No commits in common. "f6e7c8eefa58eff1b00e5e93c421fbd91b0c55dc" and "87abbbb95aeb0b72cd5b553cbec8e0901d472e4e" have entirely different histories.

13 changed files with 398 additions and 614 deletions

Binary file not shown.

View File

@ -4,7 +4,8 @@
<SchemaVersion>2.0</SchemaVersion>
<ProjectGuid>4d8c8573-32f0-4a62-9e62-3ce5cc680390</ProjectGuid>
<ProjectHome>.</ProjectHome>
<StartupFile>fame_hmm.py</StartupFile>
<StartupFile>
</StartupFile>
<SearchPath>
</SearchPath>
<WorkingDirectory>.</WorkingDirectory>
@ -22,7 +23,7 @@
</PropertyGroup>
<ItemGroup>
<Compile Include="check_novoapi.py" />
<Compile Include="convert_phoneset.py">
<Compile Include="convert_phone_set.py">
<SubType>Code</SubType>
</Compile>
<Compile Include="convert_xsampa2ipa.py">
@ -31,7 +32,7 @@
<Compile Include="defaultfiles.py">
<SubType>Code</SubType>
</Compile>
<Compile Include="fame_test.py">
<Compile Include="fame_phoneset.py">
<SubType>Code</SubType>
</Compile>
<Compile Include="fa_test.py">
@ -49,20 +50,9 @@
<SubType>Code</SubType>
</Compile>
<Compile Include="fame_hmm.py" />
<Compile Include="phoneset\fame_asr.py" />
<Compile Include="phoneset\fame_ipa.py" />
</ItemGroup>
<ItemGroup>
<Content Include="config.ini" />
<Content Include="phoneset\fame_ipa2asr.npy" />
<Content Include="phoneset\output_get_translation_key_phone_unknown.npy" />
<Content Include="phoneset\output_get_translation_key_translation_key.npy" />
<Content Include="phoneset\__pycache__\fame_asr.cpython-36.pyc" />
<Content Include="phoneset\__pycache__\fame_ipa.cpython-36.pyc" />
</ItemGroup>
<ItemGroup>
<Folder Include="phoneset\" />
<Folder Include="phoneset\__pycache__\" />
</ItemGroup>
<Import Project="$(MSBuildExtensionsPath32)\Microsoft\VisualStudio\v$(VisualStudioVersion)\Python Tools\Microsoft.PythonTools.targets" />
<!-- Uncomment the CoreCompile target to enable the Build command in

View File

@ -20,21 +20,10 @@ def split_word(word, multi_character_phones):
Args:
word (str): a word written in given phoneset.
multi_character_phones (list): the list of multicharacter phones which is considered as one phone. this can be obtained with phoneset definition such as fame_ipa.py.
multi_character_phones (list): the list of multicharacter phones which is considered as one phone. this can be obtained with phoneset definition such as fame_phoneset.py.
Returns:
(word_seperated) (list): the word splitted in given phoneset.
"""
return [phone
for phone in multi_character_tokenize(word.strip(), multi_character_phones)
]
def convert_phoneset(word_list, translation_key):
"""
Args:
word_list (str): a list of phones written in given phoneset.
translation_key (dict):
"""
return [translation_key.get(phone, phone) for phone in word_list]
return [phone for phone in multi_character_tokenize(word.strip(), multi_character_phones)]

View File

@ -1,13 +1,14 @@
import os
# add path of the parent directory
#os.path.dirname(os.path.realpath(__file__))
#default_hvite_config = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'data', 'htk', 'config.HVite')
#cygwin_dir = r'C:\cygwin64\home\Aki\acoustic_model'
#htk_dir = r'C:\Aki\htk_fame'
htk_dir = r'c:\OneDrive\Research\rug\experiments\acoustic_model\fame\htk'
config_hcopy = os.path.join(htk_dir, 'config', 'config.HCopy')
#config_train = os.path.join(cygwin_dir, 'config', 'config.train')
#config_hvite = os.path.join(cygwin_dir, 'config', 'config.HVite')
#mkhmmdefs_pl = os.path.join(cygwin_dir, 'src', 'acoustic_model', 'mkhmmdefs.pl')
@ -38,11 +39,11 @@ toolbox_dir = os.path.join(repo_dir, 'toolbox')
#config_hvite = os.path.join(htk_config_dir, 'config.HVite')
#acoustic_model = os.path.join(htk_config_dir, 'hmmdefs.compo')
#acoustic_model = r'c:\cygwin64\home\A.Kunikoshi\acoustic_model\model\barbara\hmm128-2\hmmdefs.compo'
phonelist_txt = os.path.join(htk_dir, 'config', 'phonelist.txt')
#phonelist_txt = os.path.join(htk_config_dir, 'phonelist.txt')
WSL_dir = r'C:\OneDrive\WSL'
#fame_dir = os.path.join(WSL_dir, 'kaldi-trunk', 'egs', 'fame')
fame_dir = r'c:\OneDrive\Research\rug\_data\FAME'
fame_dir = r'd:\_corpus\fame'
fame_s5_dir = os.path.join(fame_dir, 's5')
fame_corpus_dir = os.path.join(fame_dir, 'corpus')

View File

@ -9,8 +9,38 @@ import numpy as np
import pandas as pd
import defaultfiles as default
import convert_phoneset
from phoneset import fame_ipa, fame_asr
import fame_phoneset
import convert_phone_set
#def ipa2famehtk_lexicon(lexicon_file_in, lexicon_file_out):
# """ Convert a lexicon file from IPA to HTK format for FAME! corpus. """
# lexicon_in = pd.read_table(lexicon_file_in, names=['word', 'pronunciation'])
# with open(lexicon_file_out, "w", encoding="utf-8") as fout:
# for word, pronunciation in zip(lexicon_in['word'], lexicon_in['pronunciation']):
# pronunciation_no_space = pronunciation.replace(' ', '')
# pronunciation_famehtk = convert_phone_set.ipa2famehtk(pronunciation_no_space)
# if 'ceh' not in pronunciation_famehtk and 'sh' not in pronunciation_famehtk:
# fout.write("{0}\t{1}\n".format(word.upper(), pronunciation_famehtk))
#def combine_lexicon(lexicon_file1, lexicon_file2, lexicon_out):
# """ Combine two lexicon files and sort by words. """
# with open(lexicon_file1, "rt", encoding="utf-8") as fin:
# lines1 = fin.read()
# lines1 = lines1.split('\n')
# with open(lexicon_file2, "rt", encoding="utf-8") as fin:
# lines2 = fin.read()
# lines2 = lines2.split('\n')
# lex1 = pd.read_table(lexicon_file1, names=['word', 'pronunciation'])
# lex2 = pd.read_table(lexicon_file2, names=['word', 'pronunciation'])
# lex = pd.concat([lex1, lex2])
# lex = lex.sort_values(by='word', ascending=True)
# lex.to_csv(lexicon_out, index=False, header=False, encoding="utf-8", sep='\t')
#def read_fileFA(fileFA):
# """
@ -80,6 +110,14 @@ from phoneset import fame_ipa, fame_asr
# return ipa
#def make_filelist(input_dir, output_txt):
# """ Make a list of files in the input_dir. """
# filenames = os.listdir(input_dir)
# with open(output_txt, 'w') as fout:
# for filename in filenames:
# fout.write(input_dir + '\\' + filename + '\n')
#def make_htk_dict(word, pronvar_, fileDic, output_type):
# """
@ -141,11 +179,10 @@ def make_hcopy_scp_from_filelist_in_fame(fame_dir, dataset, feature_dir, hcopy_s
fout.write(wav_file + '\t' + mfc_file + '\n')
return
def load_lexicon(lexicon_file):
""" load lexicon file as data frame.
""" load lexicon file as Data Frame.
Args:
lexicon_file (path): lexicon in the format of 'word' /t 'pronunciation'.
@ -159,27 +196,25 @@ def load_lexicon(lexicon_file):
return lex
def get_phoneset_from_lexicon(lexicon_file, phoneset_name='asr'):
def get_phoneset_from_lexicon(lexicon_file, phoneset='asr'):
""" Make a list of phones which appears in the lexicon.
Args:
lexicon_file (path): lexicon in the format of 'word' /t 'pronunciation'.
phoneset_name (str): the name of phoneset with which lexicon_file is written. 'asr'(default) or 'ipa'.
phoneset (str): the phoneset with which lexicon_file is written. 'asr'(default) or 'ipa'.
Returns:
(list_of_phones) (set): the set of phones included in the lexicon_file.
"""
assert phoneset_name in ['asr', 'ipa'], 'phoneset_name should be \'asr\' or \'ipa\''
assert phoneset in ['asr', 'ipa'], 'phoneset should be \'asr\' or \'ipa\''
lex = load_lexicon(lexicon_file)
if phoneset_name == 'asr':
if phoneset == 'asr':
return set(' '.join(lex['pronunciation']).split(' '))
elif phoneset_name == 'ipa':
elif phoneset == 'ipa':
join_pronunciations = ''.join(lex['pronunciation'])
return set(convert_phone_set.split_word(join_pronunciations, fame_ipa.multi_character_phones))
return
return set(convert_phone_set.split_word(join_pronunciations, fame_phoneset.multi_character_phones_ipa))
def extract_unknown_phones(ipa, known_phones):
@ -193,7 +228,7 @@ def extract_unknown_phones(ipa, known_phones):
(list_of_phones) (list): unknown phones not included in 'known_phones'.
"""
ipa_split = convert_phone_set.split_word(ipa, fame_ipa.multi_character_phones)
ipa_split = convert_phone_set.split_word(ipa, fame_phoneset.multi_character_phones_ipa)
return [i for i in ipa_split if not i in known_phones]
@ -212,14 +247,14 @@ def get_translation_key(lexicon_file_ipa, lexicon_file_asr):
"""
lex_ipa = load_lexicon(lexicon_file_ipa)
lex_asr = load_lexicon(lexicon_file_asr)
phone_unknown = fame_ipa.phoneset[:]
phone_unknown = fame_phoneset.phoneset_ipa[:]
translation_key = dict()
for word in lex_ipa['word']:
if np.sum(lex_ipa['word'] == word) == 1 and np.sum(lex_asr['word'] == word) == 1:
ipa = lex_ipa[lex_ipa['word'] == word].iat[0, 1]
asr = lex_asr[lex_asr['word'] == word].iat[0, 1]
ipa_list = convert_phone_set.split_word(ipa, fame_ipa.multi_character_phones)
ipa_list = convert_phone_set.split_word(ipa, fame_phoneset.multi_character_phones_ipa)
asr_list = asr.split(' ')
# if there are phones which is not in phone_unknown
@ -233,13 +268,13 @@ def get_translation_key(lexicon_file_ipa, lexicon_file_asr):
return translation_key, list(phone_unknown)
def find_phone(lexicon_file, phone, phoneset_name='ipa'):
def find_phone(lexicon_file, phone, phoneset='ipa'):
""" extract rows where the phone is used in the lexicon_file.
Args:
lexicon_file (path): lexicon in the format of 'word' /t 'pronunciation'.
phone (str): the phone to be searched.
phoneset_name (str): the name of phoneset_name with which lexicon_file is written. 'asr' or 'ipa'(default).
phoneset (str): the phoneset with which lexicon_file is written. 'asr' or 'ipa'(default).
Returns:
extracted (df): rows where the phone is used.
@ -248,7 +283,7 @@ def find_phone(lexicon_file, phone, phoneset_name='ipa'):
* develop when the phonset == 'asr'.
"""
assert phoneset_name in ['asr', 'ipa'], 'phoneset_name should be \'asr\' or \'ipa\''
assert phoneset in ['asr', 'ipa'], 'phoneset should be \'asr\' or \'ipa\''
lex = load_lexicon(lexicon_file)
@ -257,87 +292,9 @@ def find_phone(lexicon_file, phone, phoneset_name='ipa'):
extracted = pd.DataFrame(index=[], columns=['word', 'pronunciation'])
for index, row in lex_.iterrows():
if phoneset_name == 'ipa':
pronunciation = convert_phone_set.split_word(row['pronunciation'], fame_ipa.multi_character_phones)
if phoneset == 'ipa':
pronunciation = convert_phone_set.split_word(row['pronunciation'], fame_phoneset.multi_character_phones_ipa)
if phone in pronunciation:
extracted_ = pd.Series([row['word'], pronunciation], index=extracted.columns)
extracted = extracted.append(extracted_, ignore_index=True)
return extracted
def asr2htk_space_delimited(pronunciation):
"""convert phoneset from asr to htk.
Args:
pronunciation (str): space delimited asr phones.
Returns:
(pronunciation) (str): space delimited asr phones in htk format (ascii).
"""
pronunciation_short = [fame_asr.reduction_key.get(i, i) for i in pronunciation.split(' ')
if not i in fame_asr.phones_to_be_removed]
return ' '.join(convert_phoneset.convert_phoneset(
pronunciation_short, fame_asr.translation_key_asr2htk))
def lexicon_asr2htk(lexicon_file_asr, lexicon_file_htk):
""" Convert a lexicon file from asr to htk format (ascii).
Args:
lexicon_file_asr (path): a lexicon file written in asr format e.g. fame/lex.asr.
lexicon_file_htk (path): a lexicon file written in htk format (ascii).
"""
lex_asr = load_lexicon(lexicon_file_asr)
def word2htk_(row):
return word2htk(row['word'])
def asr2htk_space_delimited_(row):
return asr2htk_space_delimited(row['pronunciation'])
lex_htk = pd.DataFrame({
'word': lex_asr.apply(word2htk_, axis=1).str.upper(),
'pronunciation': lex_asr.apply(asr2htk_space_delimited_, axis=1)
})
lex_htk = lex_htk.ix[:, ['word', 'pronunciation']]
lex_htk.to_csv(lexicon_file_htk, header=None, index=None, sep='\t', encoding='utf-8')
return
def combine_lexicon(lexicon_file1, lexicon_file2, lexicon_out):
""" Combine two lexicon files and sort by words.
Args:
lexicon_file1, lexicon_file2 (path): input lexicon files.
Returns:
lexicon_file_out (path): lexicon_file which lexcion_file1 and 2 are combined and sorted.
"""
lex1 = load_lexicon(lexicon_file1)
lex2 = load_lexicon(lexicon_file2)
lex = pd.concat([lex1, lex2])
lex = lex.sort_values(by='word', ascending=True)
lex.to_csv(lexicon_out, index=False, header=False, sep='\t', encoding='utf-8')
def fix_single_quote(lexicon_file):
""" add '\' before all single quote at the beginning of words.
convert special characters to ascii compatible characters.
Args:
lexicon_file (path): lexicon file, which will be overwitten.
"""
lex = load_lexicon(lexicon_file)
lex = lex.dropna() # remove N/A.
for i in lex[lex['word'].str.startswith('\'')].index.values:
lex.iat[i, 0] = lex.iat[i, 0].replace('\'', '\\\'')
# to_csv does not work with space seperator. therefore all tabs should manually be replaced.
#lex.to_csv(lexicon_file, index=False, header=False, encoding="utf-8", sep=' ', quoting=csv.QUOTE_NONE, escapechar='\\')
lex.to_csv(lexicon_file, index=False, header=False, sep='\t', encoding='utf-8')
return
def word2htk(word):
return ''.join([fame_asr.translation_key_word2htk.get(i, i) for i in word])

View File

@ -3,15 +3,15 @@ import os
os.chdir(r'C:\Users\Aki\source\repos\acoustic_model\acoustic_model')
import tempfile
import shutil
import glob
#import configparser
#import subprocess
#from collections import Counter
import time
import numpy as np
import pandas as pd
#import numpy as np
#import pandas as pd
import fame_functions
from phoneset import fame_ipa, fame_asr
import defaultfiles as default
sys.path.append(default.toolbox_dir)
import file_handling as fh
@ -19,42 +19,60 @@ from htk import pyhtk
## ======================= user define =======================
# procedure
make_lexicon = 0
make_label = 0 # it takes roughly 4800 sec on Surface pro 2.
make_htk_files = 0
extract_features = 0
flat_start = 0
train_model_without_sp = 1
# pre-defined values.
#repo_dir = 'C:\\Users\\Aki\\source\\repos\\acoustic_model'
#curr_dir = repo_dir + '\\acoustic_model'
#config_ini = curr_dir + '\\config.ini'
#output_dir = 'C:\\OneDrive\\Research\\rug\\experiments\\friesian\\acoustic_model'
#forced_alignment_module = 'C:\\Users\\Aki\\source\\repos\\forced_alignment'
dataset_list = ['devel', 'test', 'train']
hmmdefs_name = 'hmmdefs'
lexicon_asr = os.path.join(default.fame_dir, 'lexicon', 'lex.asr')
lexicon_oov = os.path.join(default.fame_dir, 'lexicon', 'lex.oov')
config_dir = os.path.join(default.htk_dir, 'config')
config_hcopy = os.path.join(config_dir, 'config.HCopy')
config_train = os.path.join(config_dir, 'config.train')
global_ded = os.path.join(config_dir, 'global.ded')
mkphones_led = os.path.join(config_dir, 'mkphones.led')
prototype = os.path.join(config_dir, 'proto39')
model_dir = os.path.join(default.htk_dir, 'model')
# procedure
extract_features = 0
conv_lexicon = 1
#check_lexicon = 0
#make_mlf = 0
#combine_files = 0
#flat_start = 0
#train_model = 1
# directories / files to be made.
#sys.path.append(os.path.join(os.path.dirname(sys.path[0]), curr_dir))
#sys.path.append(forced_alignment_module)
#from forced_alignment import convert_phone_set
lexicon_dir = os.path.join(default.htk_dir, 'lexicon')
lexicon_htk_asr = os.path.join(lexicon_dir, 'lex.htk_asr')
lexicon_htk_oov = os.path.join(lexicon_dir, 'lex.htk_oov')
lexicon_htk = os.path.join(lexicon_dir, 'lex.htk')
phonelist_txt = os.path.join(config_dir, 'phonelist.txt')
model0_dir = os.path.join(model_dir, 'hmm0')
## ======================= load variables =======================
#config = configparser.ConfigParser()
#config.sections()
#config.read(config_ini)
#config_hcopy = config['Settings']['config_hcopy']
#config_train = config['Settings']['config_train']
#mkhmmdefs_pl = config['Settings']['mkhmmdefs_pl']
#FAME_dir = config['Settings']['FAME_dir']
#lex_asr = FAME_dir + '\\lexicon\\lex.asr'
#lex_asr_htk = FAME_dir + '\\lexicon\\lex.asr_htk'
#lex_oov = FAME_dir + '\\lexicon\\lex.oov'
#lex_oov_htk = FAME_dir + '\\lexicon\\lex.oov_htk'
##lex_ipa = FAME_dir + '\\lexicon\\lex.ipa'
##lex_ipa_ = FAME_dir + '\\lexicon\\lex.ipa_'
##lex_ipa_htk = FAME_dir + '\\lexicon\\lex.ipa_htk'
#lex_htk = FAME_dir + '\\lexicon\\lex_original.htk'
#lex_htk_ = FAME_dir + '\\lexicon\\lex.htk'
#hcompv_scp = output_dir + '\\scp\\combined.scp'
#combined_mlf = output_dir + '\\label\\combined.mlf'
#model_dir = output_dir + '\\model'
#model0_dir = model_dir + '\\hmm0'
#proto_init = model_dir + '\\proto38'
#proto_name = 'proto'
#phonelist = output_dir + '\\config\\phonelist_friesian.txt'
#hmmdefs_name = 'hmmdefs'
feature_dir = os.path.join(default.htk_dir, 'mfc')
if not os.path.exists(feature_dir):
@ -62,199 +80,299 @@ if not os.path.exists(feature_dir):
tmp_dir = os.path.join(default.htk_dir, 'tmp')
if not os.path.exists(tmp_dir):
os.makedirs(tmp_dir)
label_dir = os.path.join(default.htk_dir, 'label')
if not os.path.exists(label_dir):
os.makedirs(label_dir)
## ======================= make lexicon for HTK =======================
if make_lexicon:
timer_start = time.time()
print('==== making lexicon for HTK ====')
# convert each lexicon from fame_asr phoneset to fame_htk phoneset.
print('>>> converting each lexicon from fame_asr phoneset to fame_htk phoneset...')
fame_functions.lexicon_asr2htk(lexicon_asr, lexicon_htk_asr)
fame_functions.lexicon_asr2htk(lexicon_oov, lexicon_htk_oov)
# combine lexicon
print('>>> combining lexicon files into one lexicon...')
# pronunciations which is not found in lex.asr are generated using G2P and listed in lex.oov.
# therefore there is no overlap between lex_asr and lex_oov.
fame_functions.combine_lexicon(lexicon_htk_asr, lexicon_htk_oov, lexicon_htk)
## =======================
## manually make changes to the pronunciation dictionary and save it as lex.htk
## =======================
# (1) Replace all tabs with single space;
# (2) Put a '\' before any dictionary entry beginning with single quote
#http://electroblaze.blogspot.nl/2013/03/understanding-htk-error-messages.html
fame_functions.fix_single_quote(lexicon_htk)
print("elapsed time: {}".format(time.time() - timer_start))
## ======================= make label files =======================
if make_label:
for dataset in dataset_list:
timer_start = time.time()
print("==== making label files on dataset {}".format(dataset))
script_list = os.path.join(default.fame_dir, 'data', dataset, 'text')
wav_dir_ = os.path.join(default.fame_dir, 'fame', 'wav', dataset)
label_dir_ = os.path.join(label_dir, dataset)
dictionary_file = os.path.join(label_dir_, 'temp.dic')
fh.make_new_directory(label_dir_)
# list of scripts
with open(script_list, "rt", encoding="utf-8") as fin:
scripts = fin.read().split('\n')
for line in scripts:
# sample line:
# sp0457m_test_1968_plakkenfryslanterhorne_2168 en dan begjinne je natuerlik
filename_ = line.split(' ')[0]
filename = '_'.join(filename_.split('_')[1:])
sentence = ' '.join(line.split(' ')[1:])
sentence_htk = fame_functions.word2htk(sentence)
wav_file = os.path.join(wav_dir_, filename + '.wav')
if os.path.exists(wav_file) and pyhtk.can_be_ascii(sentence_htk) == 0:
if pyhtk.create_dictionary_without_log(
sentence_htk, global_ded, dictionary_file, lexicon_htk) == 0:
# when the file name is too long, HDMan command does not work.
# therefore first temporary dictionary_file is made, then renamed.
shutil.move(dictionary_file, os.path.join(label_dir_, filename + '.dic'))
label_file = os.path.join(label_dir_, filename + '.lab')
pyhtk.create_label_file(sentence_htk, label_file)
else:
os.remove(dictionary_file)
print("elapsed time: {}".format(time.time() - timer_start))
## ======================= make other required files =======================
if make_htk_files:
timer_start = time.time()
print("==== making files required for HTK ====")
print(">>> making a phonelist...")
pyhtk.create_phonelist_file(fame_asr.phoneset_htk, phonelist_txt)
for dataset in dataset_list:
wav_dir_ = os.path.join(default.fame_dir, 'fame', 'wav', dataset)
feature_dir_ = os.path.join(feature_dir, dataset)
label_dir_ = os.path.join(label_dir, dataset)
mlf_word = os.path.join(label_dir, dataset + '_word.mlf')
mlf_phone = os.path.join(label_dir, dataset + '_phone.mlf')
#print(">>> making a script file for {}...".format(dataset))
#listdir = glob.glob(os.path.join(wav_dir_, '*.dic'))
#mfc_list = [filename.replace(wav_dir_, feature_dir_).replace('.dic', '.mfc') for filename in listdir]
#hcompv_scp = os.path.join(tmp_dir, dataset + '.scp')
#with open(hcompv_scp, 'wb') as f:
# f.write(bytes('\n'.join(mfc_list) + '\n', 'ascii'))
print(">>> making a mlf file for {}...".format(dataset))
lab_list = glob.glob(os.path.join(label_dir_, '*.lab'))
with open(mlf_word, 'wb') as fmlf:
fmlf.write(bytes('#!MLF!#\n', 'ascii'))
for label_file in lab_list:
filename = os.path.basename(label_file)
fmlf.write(bytes('\"*/{}\"\n'.format(filename), 'ascii'))
with open(label_file) as flab:
lines = flab.read()
fmlf.write(bytes(lines + '.\n', 'ascii'))
print(">>> generating phone level transcription for {}...".format(dataset))
pyhtk.mlf_word2phone(lexicon_htk, mlf_phone, mlf_word, mkphones_led)
print("elapsed time: {}".format(time.time() - timer_start))
## ======================= extract features =======================
if extract_features:
for dataset in dataset_list:
timer_start = time.time()
print('==== extract features on dataset {} ===='.format(dataset))
wav_dir_ = os.path.join(default.fame_dir, 'fame', 'wav', dataset)
label_dir_ = os.path.join(label_dir, dataset)
feature_dir_ = os.path.join(feature_dir, dataset)
fh.make_new_directory(feature_dir_)
print('==== {} ===='.format(dataset))
# a script file for HCopy
print(">>> making a script file for HCopy...")
print(">>> making a script file for HCopy... \n")
hcopy_scp = tempfile.NamedTemporaryFile(mode='w', delete=False)
hcopy_scp.close()
# get a list of features (hcopy.scp)
# from the filelist in FAME! corpus.
#fame_functions.make_hcopy_scp_from_filelist_in_fame(default.fame_dir, dataset, feature_dir_, hcopy_scp.name)
# from the list of label files.
lab_list = glob.glob(os.path.join(label_dir_, '*.lab'))
feature_list = [
os.path.join(wav_dir_, os.path.basename(lab_file).replace('.lab', '.wav')) + '\t'
+ os.path.join(feature_dir_, os.path.basename(lab_file).replace('.lab', '.mfc'))
for lab_file in lab_list]
with open(hcopy_scp.name, 'wb') as f:
f.write(bytes('\n'.join(feature_list), 'ascii'))
# get a list of features (hcopy.scp) from the filelist in FAME! corpus
feature_dir_ = os.path.join(feature_dir, dataset)
if not os.path.exists(feature_dir_):
os.makedirs(feature_dir_)
# extract features.
print(">>> extracting features on {}...".format(dataset))
pyhtk.wav2mfc(config_hcopy, hcopy_scp.name)
os.remove(hcopy_scp.name)
# extract features
print(">>> extracting features... \n")
fame_functions.make_hcopy_scp_from_filelist_in_fame(default.fame_dir, dataset, feature_dir_, hcopy_scp.name)
pyhtk.wav2mfc(default.config_hcopy, hcopy_scp.name)
# make hcompv.scp.
print(">>> making a script file for {}...".format(dataset))
listdir = glob.glob(os.path.join(label_dir_, '*.dic'))
mfc_list = [filename.replace(label_dir_, feature_dir_).replace('.dic', '.mfc') for filename in listdir]
hcompv_scp = os.path.join(tmp_dir, dataset + '.scp')
with open(hcompv_scp, 'wb') as f:
f.write(bytes('\n'.join(mfc_list) + '\n', 'ascii'))
# a script file for HCompV
print(">>> making a script file for HCompV... \n")
hcompv_scp = os.path.join(tmp_dir, dataset + '.scp')
fh.make_filelist(feature_dir_, hcompv_scp, '.mfc')
## ======================= convert lexicon from ipa to fame_htk =======================
if conv_lexicon:
print('==== convert lexicon from ipa 2 fame ====\n')
#dir_out = r'c:\Users\Aki\source\repos\acoustic_model\_tmp'
lexicon_dir = os.path.join(default.fame_dir, 'lexicon')
lexicon_ipa = os.path.join(lexicon_dir, 'lex.ipa')
lexicon_asr = os.path.join(lexicon_dir, 'lex.asr')
# get the correspondence between lex_ipa and lex_asr.
lex_asr = fame_functions.load_lexicon(lexicon_asr)
lex_ipa = fame_functions.load_lexicon(lexicon_ipa)
if 1:
timer_start = time.time()
translation_key, phone_unknown = fame_functions.get_translation_key(lexicon_ipa, lexicon_asr)
print("elapsed time: {}".format(time.time() - timer_start))
np.save('translation_key_ipa2asr.npy', translation_key)
np.save('phone_unknown.npy', phone_unknown)
else:
translation_key = np.load('translation_key_ipa2asr.npy').item()
phone_unknown = np.load('phone_unknown.npy')
phone_unknown = list(phone_unknown)
## manually check the correspondence for the phone in phone_unknown.
#p = phone_unknown[0]
#lex_ipa_ = find_phone(lexicon_ipa, p, phoneset='ipa')
#for word in lex_ipa_['word']:
# ipa = lex_ipa[lex_ipa['word'] == word].iat[0, 1]
# if np.sum(lex_asr['word'] == word) > 0:
# asr = lex_asr[lex_asr['word'] == word].iat[0, 1]
# ipa_list = convert_phone_set.split_word(ipa, fame_phoneset.multi_character_phones_ipa)
# asr_list = asr.split(' ')
# if p in ipa_list and (len(ipa_list) == len(asr_list)):
# print("{0}: {1} --> {2}".format(word, ipa_list, asr_list))
# for ipa_, asr_ in zip(ipa_list, asr_list):
# if ipa_ in phone_unknown:
# translation_key[ipa_] = asr_
# phone_unknown.remove(ipa_)
## check if all the phones in lexicon_ipa are in fame_phoneset.py.
#timer_start = time.time()
#phoneset_lex = get_phoneset_from_lexicon(lexicon_ipa, phoneset='ipa')
#print("elapsed time: {}".format(time.time() - timer_start))
#phoneset_py = fame_phoneset.phoneset_ipa
#set(phoneset_lex) - set(phoneset_py)
##timer_start = time.time()
##extracted = find_phone(lexicon_ipa, 'ⁿ')
##print("elapsed time: {}".format(time.time() - timer_start))
# lex.asr is Kaldi compatible version of lex.ipa.
# to check...
#lexicon_ipa = pd.read_table(lex_ipa, names=['word', 'pronunciation'])
#with open(lex_ipa_, "w", encoding="utf-8") as fout:
# for word, pronunciation in zip(lexicon_ipa['word'], lexicon_ipa['pronunciation']):
# # ignore nasalization and '.'
# pronunciation_ = pronunciation.replace(u'ⁿ', '')
# pronunciation_ = pronunciation_.replace('.', '')
# pronunciation_split = convert_phone_set.split_ipa_fame(pronunciation_)
# fout.write("{0}\t{1}\n".format(word, ' '.join(pronunciation_split)))
# convert each lexicon from ipa description to fame_htk phoneset.
#am_func.ipa2famehtk_lexicon(lex_oov, lex_oov_htk)
#am_func.ipa2famehtk_lexicon(lex_asr, lex_asr_htk)
# combine lexicon
# pronunciations which is not found in lex.asr are generated using G2P and listed in lex.oov.
# therefore there is no overlap between lex_asr and lex_oov.
#am_func.combine_lexicon(lex_asr_htk, lex_oov_htk, lex_htk)
## ======================= check if all the phones are successfully converted =======================
if check_lexicon:
print("==== check if all the phones are successfully converted. ====\n")
# the phones used in the lexicon.
phonelist_asr = am_func.get_phonelist(lex_asr)
phonelist_oov = am_func.get_phonelist(lex_oov)
phonelist_htk = am_func.get_phonelist(lex_htk)
phonelist = phonelist_asr.union(phonelist_oov)
# the lines which include a specific phone.
lines = am_func.find_phone(lex_asr, 'g')
# statistics over the lexicon
lexicon_htk = pd.read_table(lex_htk, names=['word', 'pronunciation'])
pronunciation = lexicon_htk['pronunciation']
phones_all = []
for word in pronunciation:
phones_all = phones_all + word.split()
c = Counter(phones_all)
## =======================
## manually make changes to the pronunciation dictionary and save it as lex.htk
## =======================
# (1) Replace all tabs with single space;
# (2) Put a '\' before any dictionary entry beginning with single quote
#http://electroblaze.blogspot.nl/2013/03/understanding-htk-error-messages.html
## ======================= make label file =======================
if make_mlf:
print("==== make mlf ====\n")
print("generating word level transcription...\n")
for dataset in dataset_list:
hcompv_scp = output_dir + '\\scp\\' + dataset + '.scp'
hcompv_scp2 = output_dir + '\\scp\\' + dataset + '_all_words_in_lexicon.scp'
script_list = FAME_dir + '\\data\\' + dataset + '\\text'
mlf_word = output_dir + '\\label\\' + dataset + '_word.mlf'
mlf_phone = output_dir + '\\label\\' + dataset + '_phone.mlf'
# lexicon
lexicon_htk = pd.read_table(lex_htk, names=['word', 'pronunciation'])
# list of features
with open(hcompv_scp) as fin:
features = fin.read()
features = features.split('\n')
# list of scripts
with open(script_list, "rt", encoding="utf-8") as fin:
scripts = fin.read()
scripts = pd.Series(scripts.split('\n'))
i = 0
missing_words = []
fscp = open(hcompv_scp2, 'wt')
fmlf = open(mlf_word, "wt", encoding="utf-8")
fmlf.write("#!MLF!#\n")
feature_nr = 1
for feature in features:
sys.stdout.write("\r%d/%d" % (feature_nr, len(features)))
sys.stdout.flush()
feature_nr += 1
file_basename = os.path.basename(feature).replace('.mfc', '')
# get words from scripts.
try:
script = scripts[scripts.str.contains(file_basename)]
except IndexError:
script = []
if len(script) != 0:
script_id = script.index[0]
script_txt = script.get(script_id)
script_words = script_txt.split(' ')
del script_words[0]
# check if all words can be found in the lexicon.
SCRIPT_WORDS = []
script_prons = []
is_in_lexicon = 1
for word in script_words:
WORD = word.upper()
SCRIPT_WORDS.append(WORD)
extracted = lexicon_htk[lexicon_htk['word']==WORD]
if len(extracted) == 0:
missing_words.append(word)
script_prons.append(extracted)
is_in_lexicon *= len(extracted)
# if all pronunciations are found in the lexicon, update scp and mlf files.
if is_in_lexicon:
# add the feature filename into the .scp file.
fscp.write("{}\n".format(feature))
i += 1
# add the words to the mlf file.
fmlf.write('\"*/{}.lab\"\n'.format(file_basename))
#fmlf.write('{}'.format('\n'.join(SCRIPT_WORDS)))
for word_ in SCRIPT_WORDS:
if word_[0] == '\'':
word_ = '\\' + word_
fmlf.write('{}\n'.format(word_))
fmlf.write('.\n')
print("\n{0} has {1} samples.\n".format(dataset, i))
np.save(output_dir + '\\missing_words' + '_' + dataset + '.npy', missing_words)
fscp.close()
fmlf.close()
## generate phone level transcription
print("generating phone level transcription...\n")
mkphones = output_dir + '\\label\\mkphones0.txt'
subprocessStr = r"HLEd -l * -d " + lex_htk_ + ' -i ' + mlf_phone + ' ' + mkphones + ' ' + mlf_word
subprocess.call(subprocessStr, shell=True)
## ======================= combined scps and mlfs =======================
if combine_files:
print("==== combine scps and mlfs ====\n")
fscp = open(hcompv_scp, 'wt')
fmlf = open(combined_mlf, 'wt')
for dataset in dataset_list:
fmlf.write("#!MLF!#\n")
for dataset in dataset_list:
each_mlf = output_dir + '\\label\\' + dataset + '_phone.mlf'
each_scp = output_dir + '\\scp\\' + dataset + '_all_words_in_lexicon.scp'
with open(each_mlf, 'r') as fin:
lines = fin.read()
lines = lines.split('\n')
fmlf.write('\n'.join(lines[1:]))
with open(each_scp, 'r') as fin:
lines = fin.read()
fscp.write(lines)
fscp.close()
fmlf.close()
## ======================= flat start monophones =======================
if flat_start:
hcompv_scp = os.path.join(tmp_dir, 'test.scp')
timer_start = time.time()
print('==== flat start ====')
pyhtk.flat_start(config_train, hcompv_scp, model0_dir, prototype)
subprocessStr = 'HCompV -T 1 -C ' + config_train + ' -m -v 0.01 -S ' + hcompv_scp + ' -M ' + model0_dir + ' ' + proto_init
subprocess.call(subprocessStr, shell=True)
# allocate mean & variance to all phones in the phone list
pyhtk.create_hmmdefs(
os.path.join(model0_dir, 'proto39'),
os.path.join(model0_dir, 'hmmdefs'),
phonelist_txt)
print("elapsed time: {}".format(time.time() - timer_start))
subprocessStr = 'perl ' + mkhmmdefs_pl + ' ' + model0_dir + '\\proto38' + ' ' + phonelist + ' > ' + model0_dir + '\\' + hmmdefs_name
subprocess.call(subprocessStr, shell=True)
## ======================= estimate monophones =======================
if train_model_without_sp:
hcompv_scp = os.path.join(tmp_dir, 'test.scp')
mlf_file = os.path.join(label_dir, 'test_phone.mlf')
output_dir = os.path.join(model_dir, 'hmm1')
fh.make_new_directory(output_dir)
if train_model:
iter_num_max = 3
for mix_num in [128, 256, 512, 1024]:
for iter_num in range(1, iter_num_max+1):
print("===== mix{}, iter{} =====".format(mix_num, iter_num))
iter_num_pre = iter_num - 1
modelN_dir = model_dir + '\\hmm' + str(mix_num) + '-' + str(iter_num)
if not os.path.exists(modelN_dir):
os.makedirs(modelN_dir)
print('==== train model without sp ====')
if not os.path.exists(os.path.join(output_dir, 'iter0')):
shutil.copytree(model0_dir, os.path.join(output_dir, 'iter0'))
niter = 1
for niter in range(1, 5):
timer_start = time.time()
hmm_n = 'iter' + str(niter)
hmm_n_pre = 'iter' + str(niter-1)
modeln_dir = os.path.join(output_dir, hmm_n)
modeln_dir_pre = os.path.join(output_dir, hmm_n_pre)
if iter_num == 1 and mix_num == 1:
modelN_dir_pre = model0_dir
else:
modelN_dir_pre = model_dir + '\\hmm' + str(mix_num) + '-' + str(iter_num_pre)
## re-estimation
subprocessStr = 'HERest -T 1 -C ' + config_train + ' -v 0.01 -I ' + combined_mlf + ' -H ' + modelN_dir_pre + '\\' + hmmdefs_name + ' -M ' + modelN_dir + ' ' + phonelist + ' -S ' + hcompv_scp
subprocess.call(subprocessStr, shell=True)
mix_num_next = mix_num * 2
modelN_dir_next = model_dir + '\\hmm' + str(mix_num_next) + '-0'
if not os.path.exists(modelN_dir_next):
os.makedirs(modelN_dir_next)
header_file = modelN_dir + '\\mix' + str(mix_num_next) + '.hed'
with open(header_file, 'w') as fout:
fout.write("MU %d {*.state[2-4].mix}" % (mix_num_next))
subprocessStr = 'HHEd -T 1 -H ' + modelN_dir + '\\' + hmmdefs_name + ' -M ' + modelN_dir_next + ' ' + header_file + ' ' + phonelist
subprocess.call(subprocessStr, shell=True)
# re-estimation
fh.make_new_directory(modeln_dir)
pyhtk.re_estimation(
config_train,
os.path.join(modeln_dir_pre, 'proto39'),
os.path.join(modeln_dir_pre, hmmdefs_name),
modeln_dir,
hcompv_scp, phonelist_txt,
mlf_file=mlf_file)
print("elapsed time: {}".format(time.time() - timer_start))

View File

@ -1,6 +1,7 @@
""" definition of the phones to be used. """
phoneset = [
## phones in IPA.
phoneset_ipa = [
# vowels
'',
'i̯ⁿ',
@ -34,7 +35,7 @@ phoneset = [
'',
'ṷ.',
'ṷⁿ',
#'ú', # only appears in word 'feeste'(út) and 'gaste'(út) which are 'f e: s t ə' and 'yn' in lex_asr. The pronunciation in Fries may be mistakes so I removed this phone.
#'ú', # only appears in word 'feeste'(út) and 'gaste'(út) which are 'f e: s t ə' and 'yn' in lex_asr.
'u',
'uⁿ',
'u.',
@ -100,8 +101,7 @@ phoneset = [
'l'
]
## the list of multi character phones.
# for example, the length of 'i̯ⁿ' is 3, but in the codes it is treated as one letter.
multi_character_phones = [i for i in phoneset if len(i) > 1]
multi_character_phones.sort(key=len, reverse=True)
multi_character_phones_ipa = [i for i in phoneset_ipa if len(i) > 1]
multi_character_phones_ipa.sort(key=len, reverse=True)

View File

@ -1,134 +0,0 @@
import sys
import os
os.chdir(r'C:\Users\Aki\source\repos\acoustic_model\acoustic_model')
from collections import Counter
import time
import re
import numpy as np
import pandas as pd
import fame_functions
import defaultfiles as default
sys.path.append(default.toolbox_dir)
from phoneset import fame_ipa, fame_asr
import convert_phoneset
lexicon_dir = os.path.join(default.fame_dir, 'lexicon')
lexicon_ipa = os.path.join(lexicon_dir, 'lex.ipa')
lexicon_asr = os.path.join(lexicon_dir, 'lex.asr')
lexicon_htk = os.path.join(default.htk_dir, 'lexicon', 'lex.htk')
## check if all the phones in lexicon.ipa are in fame_ipa.py.
#timer_start = time.time()
#phoneset_lex = fame_functions.get_phoneset_from_lexicon(lexicon_ipa, phoneset='ipa')
#phoneset_py = fame_ipa.phoneset
#print("phones which is in lexicon.ipa but not in fame_ipa.py:\n{}".format(
# set(phoneset_lex) - set(phoneset_py)))
#print("elapsed time: {}".format(time.time() - timer_start))
# check which word has the phone.
#timer_start = time.time()
#extracted = find_phone(lexicon_ipa, 'ⁿ')
#print("elapsed time: {}".format(time.time() - timer_start))
## get the correspondence between lex_ipa and lex_asr.
lex_asr = fame_functions.load_lexicon(lexicon_asr)
lex_ipa = fame_functions.load_lexicon(lexicon_ipa)
if 0:
timer_start = time.time()
translation_key_ipa2asr, phone_unknown = fame_functions.get_translation_key(lexicon_ipa, lexicon_asr)
print("elapsed time: {}".format(time.time() - timer_start))
np.save(os.path.join('phoneset', 'output_get_translation_key_translation_key.npy'), translation_key_ipa2asr)
np.save(os.path.join('phoneset', 'output_get_translation_key_phone_unknown.npy'), phone_unknown)
else:
translation_key_ipa2asr = np.load(os.path.join('phoneset', 'output_get_translation_key_translation_key.npy')).item()
phone_unknown = np.load(os.path.join('phoneset', 'output_get_translation_key_phone_unknown.npy'))
phone_unknown = list(phone_unknown)
# manually check the correspondence for the phone in phone_unknown.
#p = phone_unknown[0]
#lex_ipa_ = find_phone(lexicon_ipa, p, phoneset='ipa')
#for word in lex_ipa_['word']:
# ipa = lex_ipa[lex_ipa['word'] == word].iat[0, 1]
# if np.sum(lex_asr['word'] == word) > 0:
# asr = lex_asr[lex_asr['word'] == word].iat[0, 1]
# ipa_list = convert_phone_set.split_word(ipa, fame_ipa.multi_character_phones)
# asr_list = asr.split(' ')
# if p in ipa_list and (len(ipa_list) == len(asr_list)):
# print("{0}: {1} --> {2}".format(word, ipa_list, asr_list))
# for ipa_, asr_ in zip(ipa_list, asr_list):
# if ipa_ in phone_unknown:
# translation_key_ipa2asr[ipa_] = asr_
# phone_unknown.remove(ipa_)
translation_key_ipa2asr['ə:'] = 'ə'
translation_key_ipa2asr['r.'] = 'r'
translation_key_ipa2asr['r:'] = 'r'
np.save(os.path.join('phoneset', 'fame_ipa2asr.npy'), translation_key_ipa2asr)
## check if all the phones in lexicon.asr are in translation_key_ipa2asr.
#timer_start = time.time()
#phoneset_lex = fame_functions.get_phoneset_from_lexicon(lexicon_asr, phoneset='asr')
#phoneset_lex.remove("")
#phoneset_asr = list(set(translation_key_ipa2asr.values()))
#print("phones which is in lexicon.asr but not in the translation_key_ipa2asr:\n{}".format(
# set(phoneset_lex) - set(phoneset_asr)))
#print("elapsed time: {}".format(time.time() - timer_start))
## check if all the phones in lexicon.htk are in fame_asr.py.
#timer_start = time.time()
#phoneset_htk = fame_asr.phoneset_htk
#phoneset_lex = fame_functions.get_phoneset_from_lexicon(lexicon_htk)
#phoneset_lex.remove('')
#print("phones which is in lexicon.htk but not in the fame_asr.py are:\n{}".format(
# set(phoneset_htk) - set(phoneset_lex)))
#print("elapsed time: {}".format(time.time() - timer_start))
## statistics over the lexicon
#lex_htk = fame_functions.load_lexicon(lexicon_htk)
#phones_all = (' '.join(lex_htk['pronunciation'])).split(' ')
#c = Counter(phones_all)
#lexicon_out = r'c:\OneDrive\Research\rug\experiments\acoustic_model\fame\htk\lexicon\lex.htk2'
#for i in lex_htk[lex_htk['word'].str.startswith('\'')].index.values:
# lex_htk.iat[i, 0] = lex_htk.iat[i, 0].replace('\'', '\\\'')
## to_csv does not work with space seperator. therefore all tabs should manually be replaced.
##lex_htk.to_csv(lexicon_out, index=False, header=False, encoding="utf-8", sep=' ', quoting=csv.QUOTE_NONE, escapechar='\\')
#lex_htk.to_csv(lexicon_out, index=False, header=False, encoding="utf-8", sep='\t')
## check which letters are not coded in ascii.
print('asr phones which cannot be coded in ascii:\n')
for i in fame_asr.phoneset_short:
try:
i_encoded = i.encode("ascii")
#print("{0} --> {1}".format(i, i.encode("ascii")))
except UnicodeEncodeError:
print(">>> {}".format(i))
print("letters in the scripts which is not coded in ascii:\n")
for dataset in ['train', 'devel', 'test']:
timer_start = time.time()
script_list = os.path.join(default.fame_dir, 'data', dataset, 'text')
with open(script_list, "rt", encoding="utf-8") as fin:
scripts = fin.read().split('\n')
for line in scripts:
sentence = ' '.join(line.split(' ')[1:])
sentence_htk = fame_functions.word2htk(sentence)
#if len(re.findall(r'[âêôûč\'àéèúćäëïöü]', sentence))==0:
try:
sentence_htk = bytes(sentence_htk, 'ascii')
except UnicodeEncodeError:
print(sentence)
print(sentence_htk)

View File

@ -1,137 +0,0 @@
""" definition of the phones to be used. """
# phonese in {FAME}/lexicon/lex.asr
phoneset = [
# vowels
'a',
'a:',
'e',
'e:',
'i',
'i:',
'',
'o',
'o:',
'ö',
'ö:',
'u',
'u:',
'ü',
'ü:',
#'ú', # only appears in word 'feeste'(út) and 'gaste'(út) which are 'f e: s t ə' and 'yn' in lex_asr. The pronunciation in Fries may be mistakes so I removed this phone.
'',
'y',
'ɔ',
'ɔ:',
'ɔ̈',
'ɔ̈:',
'ə',
'ɛ',
'ɛ:',
'ɪ',
'ɪ:',
# plosives
'p',
'b',
't',
'd',
'k',
'g',
'ɡ', # = 'g'
# nasals
'm',
'n',
'ŋ',
# fricatives
'f',
'v',
's',
's:',
'z',
'x',
'h',
# tap and flip
'r',
'r:',
# approximant
'j',
'l'
]
## reduce the number of phones.
# the phones which seldom occur are replaced with another more popular phones.
# replacements are based on the advice from Martijn Wieling.
reduction_key = {
'y':'i:', 'e':'e:', 'ə:':'ɛ:', 'r:':'r', 'ɡ':'g'
}
# already removed beforehand in phoneset. Just to be sure.
phones_to_be_removed = ['ú', 's:', 'ɔ̈:']
phoneset_short = [reduction_key.get(i, i) for i in phoneset
if not i in phones_to_be_removed]
phoneset_short = list(set(phoneset_short))
phoneset_short.sort()
## translation_key to htk format (ascii).
# phones which gives UnicodeEncodeError when phone.encode("ascii")
# are replaced with other characters.
translation_key_asr2htk = {
'': 'i_',
'': 'u_',
# on the analogy of German umlaut, 'e' is used.
'ö': 'oe', 'ö:': 'oe:',
'ü': 'ue', 'ü:': 'ue:',
# on the analogy of Chinese...
'ŋ': 'ng',
# refer to Xsampa.
'ɔ': 'O', 'ɔ:': 'O:', 'ɔ̈': 'Oe',
'ɛ': 'E', 'ɛ:': 'E:',
'ɪ': 'I', 'ɪ:': 'I:',
# it is @ in Xsampa, but that is not handy on HTK.
'ə': 'A'
}
phoneset_htk = [translation_key_asr2htk.get(i, i) for i in phoneset_short]
#not_in_ascii = [
# '\'',
# 'â', 'ê', 'ô', 'û', 'č',
# 'à', 'í', 'é', 'è', 'ú', 'ć',
# 'ä', 'ë', 'ï', 'ö', 'ü'
#]
translation_key_word2htk = {
#'\'': '\\\'',
'í':'i1', 'é':'e1', 'ú':'u1', 'ć':'c1',
'à':'a2', 'è':'e2',
'â':'a3', 'ê':'e3', 'ô':'o3', 'û':'u3',
'č':'c4',
'ä': 'ao', 'ë': 'ee', 'ï': 'ie', 'ö': 'oe', 'ü': 'ue',
}
#[translation_key_word2htk.get(i, i) for i in not_in_ascii]
## the list of multi character phones.
# for example, the length of 'a:' is 3, but in the codes it is treated as one letter.
# original.
multi_character_phones = [i for i in phoneset if len(i) > 1]
multi_character_phones.sort(key=len, reverse=True)
# phonset reduced.
multi_character_phones_short = [i for i in phoneset_short if len(i) > 1]
multi_character_phones_short.sort(key=len, reverse=True)
# htk compatible.
multi_character_phones_htk = [i for i in phoneset_htk if len(i) > 1]
multi_character_phones_htk.sort(key=len, reverse=True)