correspondence between lex.asr and lex.ipa is automatically obtained. header is added to the functions in fame_functions.py.

acoustic_model/convert_phone_set.py

@@ -16,14 +16,14 @@ def multi_character_tokenize(line, multi_character_tokens):
 
 def split_word(word, multi_character_phones):
 	"""
-	Split a line by given phoneset.
+	split a line by given phoneset.
 	
 	Args:
-		word (str): one word written in given phoneset.
+		word (str): a word written in given phoneset.
-		multi_character_phones:
+		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 (str): the word splitted in given phoneset. 
+		(word_seperated) (list): the word splitted in given phoneset. 
-	"""
 
+	"""
 	return [phone for phone in multi_character_tokenize(word.strip(), multi_character_phones)]

acoustic_model/defaultfiles.py

@@ -33,7 +33,6 @@ repo_dir = r'C:\Users\Aki\source\repos'
 ipa_xsampa_converter_dir    = os.path.join(repo_dir, 'ipa-xsama-converter')
 forced_alignment_module_dir = os.path.join(repo_dir, 'forced_alignment')
 accent_classification_dir   = os.path.join(repo_dir, 'accent_classification', 'accent_classification')
-#pyhtk_dir                   = os.path.join(repo_dir, 'pyhtk', 'pyhtk')
 toolbox_dir					= os.path.join(repo_dir, 'toolbox')
 
 #htk_config_dir = r'c:\Users\A.Kunikoshi\source\repos\forced_alignment\forced_alignment\data\htk\preset_models\aki_dutch_2017'

acoustic_model/fame_functions.py

@@ -12,24 +12,6 @@ import defaultfiles as default
 import fame_phoneset
 import convert_phone_set
 
-#sys.path.append(default.forced_alignment_module_dir)
-#from forced_alignment import convert_phone_set
-
-#def find_phone(lexicon_file, phone):
-#	""" Search where the phone is used in the lexicon. """
-#	with open(lexicon_file, "rt", encoding="utf-8") as fin:
-#		lines = fin.read()
-#		lines = lines.split('\n')
-	
-#	extracted = []
-#	for line in lines:
-#		line = line.split('\t')
-#		if len(line) > 1:
-#			pronunciation = line[1]
-#			if phone in pronunciation:
-#				extracted.append(line)
-#	return extracted
-
 
 #def ipa2famehtk_lexicon(lexicon_file_in, lexicon_file_out):
 #	""" Convert a lexicon file from IPA to HTK format for FAME! corpus. """
@@ -128,25 +110,6 @@ import convert_phone_set
 
 #    return ipa
 
-def make_hcopy_scp_from_filelist_in_fame(fame_dir, dataset, feature_dir, hcopy_scp):
-	""" Make a script file for HCopy using the filelist in FAME! corpus. """
-	
-	filelist_txt = os.path.join(fame_dir, 'fame', 'filelists', dataset + 'list.txt')
-	with open(filelist_txt) as fin:
-		filelist = fin.read()
-		filelist = filelist.split('\n')
-	
-	with open(hcopy_scp, 'w') as fout:
-		for filename_ in filelist:
-			filename = filename_.replace('.TextGrid', '')
-
-			if len(filename) > 3: # remove '.', '..' and ''
-				wav_file = os.path.join(fame_dir, 'fame', 'wav', dataset, filename + '.wav')
-				mfc_file = os.path.join(feature_dir, filename + '.mfc')
-
-				fout.write(wav_file + '\t' + mfc_file + '\n')
-
-
 #def make_filelist(input_dir, output_txt):
 #	""" Make a list of files in the input_dir. """
 #	filenames = os.listdir(input_dir)
@@ -191,48 +154,100 @@ def make_hcopy_scp_from_filelist_in_fame(fame_dir, dataset, feature_dir, hcopy_s
 #                            f.write('{0}\t{1}\n'.format(WORD, key))
 
 
+def make_hcopy_scp_from_filelist_in_fame(fame_dir, dataset, feature_dir, hcopy_scp):
+	""" Make a script file for HCopy using the filelist in FAME! corpus. 
+	
+	Args:
+		fame_dir (path): the directory of FAME corpus.
+		dataset (str): 'devel', 'test' or 'train'.
+		feature_dir (path): the directory where feature will be stored.
+		hcopy_scp (path): a script file for HCopy to be made.
+
+	"""
+	filelist_txt = os.path.join(fame_dir, 'fame', 'filelists', dataset + 'list.txt')
+	with open(filelist_txt) as fin:
+		filelist = fin.read()
+		filelist = filelist.split('\n')
+	
+	with open(hcopy_scp, 'w') as fout:
+		for filename_ in filelist:
+			filename = filename_.replace('.TextGrid', '')
+
+			if len(filename) > 3: # remove '.', '..' and ''
+				wav_file = os.path.join(fame_dir, 'fame', 'wav', dataset, filename + '.wav')
+				mfc_file = os.path.join(feature_dir, filename + '.mfc')
+
+				fout.write(wav_file + '\t' + mfc_file + '\n')
 
 
 
 def load_lexicon(lexicon_file):
+	""" load lexicon file as Data Frame.
+
+	Args:
+		lexicon_file (path): lexicon in the format of 'word' /t 'pronunciation'.
+	
+	Returns:
+		lex (df): lexicon as Data Frame, which has columns 'word' and 'pronunciation'.
+
+	"""
 	lex = pd.read_csv(lexicon_file, delimiter='\t', header=None, encoding="utf-8")
 	lex.rename(columns={0: 'word', 1: 'pronunciation'}, inplace=True)
 	return lex
 
 
-def get_phonelist(lexicon_asr):
+def get_phoneset_from_lexicon(lexicon_file, phoneset='asr'):
-	""" Make a list of phones which appears in the lexicon. """
+	""" Make a list of phones which appears in the lexicon. 
 	
-	#with open(lexicon_file, "rt", encoding="utf-8") as fin:
+	Args:
-	#	lines = fin.read()
+		lexicon_file (path): lexicon in the format of 'word' /t 'pronunciation'.
-	#	lines = lines.split('\n')
+		phoneset (str): the phoneset with which lexicon_file is written. 'asr'(default) or 'ipa'.
-	#	phonelist = set([])
+
-	#	for line in lines:
+	Returns:
-	#		line = line.split('\t')
+		(list_of_phones) (set): the set of phones included in the lexicon_file.
-	#		if len(line) > 1:
+
-	#			pronunciation = set(line[1].split())
+	"""
-	#			phonelist = phonelist | pronunciation
+	assert phoneset in ['asr', 'ipa'], 'phoneset should be \'asr\' or \'ipa\''
-	lex = load_lexicon(lexicon_asr)
+
+	lex = load_lexicon(lexicon_file)
+	if phoneset == 'asr':
 		return set(' '.join(lex['pronunciation']).split(' '))
+	elif phoneset == 'ipa':
+		join_pronunciations = ''.join(lex['pronunciation'])
+		return set(convert_phone_set.split_word(join_pronunciations, fame_phoneset.multi_character_phones_ipa))
 
 
-def extract_unknown_phones(word_list, known_phones):
+def extract_unknown_phones(ipa, known_phones):
-	return [i for i in word_list if not i in known_phones]
+	"""extract unknown phones in the pronunciation written in IPA.
+
+	Args:
+		ipa (str): a pronunciation written in IPA. 
+		known_phones (list): list of phones already know.
+
+	Returns:
+		(list_of_phones) (list): unknown phones not included in 'known_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]
 
 
-if __name__ == '__main__':
+def get_translation_key(lexicon_file_ipa, lexicon_file_asr):
-	import time
+	""" get correspondence between lexicon_file_ipa and lexicon_file_asr.
-	timer_start = time.time()
 
-	#def get_translation_key():
+	Args:
-	dir_tmp = r'c:\Users\Aki\source\repos\acoustic_model\_tmp'
+		lexicon_file_ipa (path): lexicon in the format of 'word' /t 'pronunciation (IPA)'.
-	lexicon_ipa = r'd:\_corpus\FAME\lexicon\lex.ipa'
+		lexicon_file_asr (path): lexicon in the format of 'word' /t 'pronunciation (asr)'.
-	lexicon_asr = r'd:\_corpus\FAME\lexicon\lex.asr'
+			the each character of 'pronunciation' should be delimited by ' '.
 
-	lex_ipa = load_lexicon(lexicon_ipa)
+	Returns:
-	lex_asr = load_lexicon(lexicon_asr)
+		translation_key (dict): translation key from ipa to asr. 
-	if 1:
+		(phone_unknown) (list): the list of IPA phones, which does not appear in lexicon_file_asr. 
-		phone_to_be_searched = fame_phoneset.phoneset_ipa[:]
+
+	"""
+	lex_ipa = load_lexicon(lexicon_file_ipa)
+	lex_asr = load_lexicon(lexicon_file_asr)
+	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:
@@ -242,47 +257,44 @@ if __name__ == '__main__':
 			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_to_be_searched
+			# if there are phones which is not in phone_unknown
-				#if len([True for i in asr_list if i in phone_to_be_searched]) > 0:
+			#if len([True for i in asr_list if i in phone_unknown]) > 0:
 			if(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_to_be_searched:
+					if ipa_ in phone_unknown:
 						translation_key[ipa_] = asr_
-							phone_to_be_searched.remove(ipa_)
+						phone_unknown.remove(ipa_)
-
+	return translation_key, list(phone_unknown)
-		print("elapsed time: {}".format(time.time() - timer_start))
-
-		np.save(os.path.join(dir_tmp, 'translation_key.npy'), translation_key)
-		np.save(os.path.join(dir_tmp, 'phone_to_be_searched.npy'), phone_to_be_searched)
-	else:
-		translation_key		 = np.load(os.path.join(dir_tmp, 'translation_key.npy')).item()
-		phone_to_be_searched = np.load(os.path.join(dir_tmp, 'phone_to_be_searched.npy')).item()
 
 
-	#phone_unknown = list(phone_to_be_searched)
+def find_phone(lexicon_file, phone, phoneset='ipa'):
-	##phone_unknown.remove('')
+	""" extract rows where the phone is used in the lexicon_file. 
-	#phone_known = list(translation_key.keys())
 
-	#p = phone_unknown[0]
+	Args:
+		lexicon_file (path): lexicon in the format of 'word' /t 'pronunciation'.
+		phone (str): the phone to be searched.
+		phoneset (str): the phoneset with which lexicon_file is written. 'asr' or 'ipa'(default).
 
-	### extract lines which contains 'unknown' phone.
+	Returns:
-	#lex_ipa_ = lex_ipa[lex_ipa['pronunciation'].str.count(p)>0]
+		extracted (df): rows where the phone is used.
-	##phone_unknown_ = phone_unknown[:]
-	##phone_unknown_.remove(p)
-	#phone_known_ = phone_known[:]
-	#phone_known_.append(p)
-	#for index, row in lex_ipa_.iterrows():
-	#	ipa = row['pronunciation']
-	#	phone_extract_unknown_phones(asr_list, phone_known_):
 
-	#	# check the number of phones in phone_unknown_
+	ToDo:
-	#	if len([True for i in asr_list if i in phone_unknown_]) == 0:
+		* develop when the phonset == 'asr'.
-	#		word = row['word']
-	#		ipa = lex_ipa[lex_ipa['word'] == word].iat[0, 1]
-	#		print("{0}: {1} --> {2}".format(word, ipa, asr))
-	#		#print("{0}:{1}".format(index, row['pronunciation']))
 
+	"""
+	assert phoneset in ['asr', 'ipa'], 'phoneset should be \'asr\' or \'ipa\''
 	
+	lex = load_lexicon(lexicon_file)
 	
+	# to reduce the calculation time, only target rows which include 'phone' at least once. 
+	lex_ = lex[lex['pronunciation'].str.count(phone)>0]
 
+	extracted = pd.DataFrame(index=[], columns=['word', 'pronunciation'])
+	for index, row in lex_.iterrows():
+		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

acoustic_model/fame_hmm.py

@@ -6,6 +6,7 @@ import tempfile
 #import configparser
 #import subprocess
 #from collections import Counter
+import time
 
 #import numpy as np
 #import pandas as pd
@@ -27,8 +28,8 @@ from htk import pyhtk
 dataset_list = ['devel', 'test', 'train']
 
 # procedure
-extract_features  = 1
+extract_features  = 0
-#conv_lexicon	  = 0
+conv_lexicon	  = 1
 #check_lexicon	  = 0
 #make_mlf		  = 0
 #combine_files	  = 0
@@ -84,16 +85,14 @@ if not os.path.exists(tmp_dir):
 ## ======================= extract features =======================
 if extract_features:
 	for dataset in dataset_list:
-	#for dataset in ['test']:
 		print('==== {} ===='.format(dataset))
 
 		# a script file for HCopy 
 		print(">>> making a script file for HCopy... \n")
 		hcopy_scp = tempfile.NamedTemporaryFile(mode='w', delete=False)
 		hcopy_scp.close()
-		#hcopy_scp = os.path.join(default.htk_dir, 'tmp', 'HCopy.scp')
 
-		## get a list of features (hcopy.scp) from the filelist in FAME! corpus
+		# 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_)
@@ -101,26 +100,11 @@ if extract_features:
 		# extract features
 		print(">>> extracting features... \n")
 		fame_functions.make_hcopy_scp_from_filelist_in_fame(default.fame_dir, dataset, feature_dir_, hcopy_scp.name)
-
-		#subprocessStr = 'HCopy -C ' + config_hcopy + ' -S ' + hcopy_scp.name
-		#subprocess.call(subprocessStr, shell=True)
 		pyhtk.wav2mfc(default.config_hcopy, hcopy_scp.name)
 
 		# a script file for HCompV
 		print(">>> making a script file for HCompV... \n")
-
-
-## ======================= make a list of features =======================
-#if make_feature_list:
-#	print("==== make a list of features ====\n")
-
-#	for dataset in dataset_list:
-#		print(dataset)
-
-		#feature_dir = output_dir + '\\mfc\\' + dataset
 		hcompv_scp  = os.path.join(tmp_dir, dataset + '.scp')
-
-		#am_func.make_filelist(feature_dir, hcompv_scp)
 		fh.make_filelist(feature_dir_, hcompv_scp, '.mfc')
 
 
@@ -128,6 +112,59 @@ if extract_features:
 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'])
@@ -140,13 +177,13 @@ if conv_lexicon:
 	#		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_oov, lex_oov_htk)
-	am_func.ipa2famehtk_lexicon(lex_asr, lex_asr_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)
+	#am_func.combine_lexicon(lex_asr_htk, lex_oov_htk, lex_htk)
 
 
 ## ======================= check if all the phones are successfully converted =======================

acoustic_model/fame_phoneset.py

@@ -1,41 +1,79 @@
+""" definition of the phones to be used. """
+
+## phones in IPA.
 phoneset_ipa = [
 	# vowels
 	'i̯',
+	'i̯ⁿ',
 	'y',
 	'i',
+	'i.',
+	'iⁿ',
 	'i:',
+	'i:ⁿ',
 	'ɪ',
-	'ɪ:',
+	'ɪⁿ',
+	'ɪ.',
+	#'ɪ:', # not included in lex.ipa
+	'ɪ:ⁿ',
 	'e',
 	'e:',
+	'e:ⁿ',
 	'ə',
+	'əⁿ',
 	'ə:',
 	'ɛ',
+	'ɛ.',
+	'ɛⁿ',
 	'ɛ:',
+	'ɛ:ⁿ',
 	'a',
+	'aⁿ',
+	'a.',
 	'a:',
+	'a:ⁿ',
 	'ṷ',
-	'ú',
+	'ṷ.',
+	'ṷⁿ',
+	#'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.',
 	'u:',
+	'u:ⁿ',
 	'ü',
+	'ü.',
+	'üⁿ',
 	'ü:',
+	'ü:ⁿ',
 	'o',
+	'oⁿ',
+	'o.',
 	'o:',
+	'o:ⁿ',
 	'ö',
+	'ö.',
+	'öⁿ',
 	'ö:',
+	'ö:ⁿ',
 	'ɔ',
+	'ɔ.',
+	'ɔⁿ',
 	'ɔ:',
-	'ɔ̈',
+	'ɔ:ⁿ',
+	#'ɔ̈', # not included in lex.ipa 
+	'ɔ̈.',
 	'ɔ̈:',
 
 	# plosives
 	'p', 
 	'b', 
 	't',
+	'tⁿ',
 	'd', 
 	'k',
 	'g',
+	'ɡ', # = 'g'
 
 	# nasals
 	'm',
@@ -48,8 +86,22 @@ phoneset_ipa = [
 	's',
 	's:',
 	'z',
+	'zⁿ',
 	'x',
 	'h',
+
+	# tap and flip
+	'r',
+	'r.', # only appears in word 'mearpartijestelsel'(does not exist in lex_asr) and 'tenoarpartij'.   
+	'r:', # only appears in word 'mûsearflearmûs' and 'sjochdêr'.
+
+	# approximant
+	'j',
+	'j.',
+	'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_ipa = [i for i in phoneset_ipa if len(i) > 1]
+multi_character_phones_ipa.sort(key=len, reverse=True)