def test_language_switch():
text = '\n'.join([
"j'aime l'anglais", "j'aime le football", "football",
"surtout le real madrid", "n'utilise pas google"
])
backend = EspeakBackend('fr-fr', language_switch='keep-flags')
out = backend._phonemize_aux(text, separator.Separator(), True)
assert out == [
'ʒɛm lɑ̃ɡlɛ', 'ʒɛm lə- (en)fʊtbɔːl(fr)', '(en)fʊtbɔːl(fr)',
'syʁtu lə- (en)ɹiəl(fr) madʁid', 'nytiliz pa (en)ɡuːɡəl(fr)'
]
# default behavior is to keep the flags
backend = EspeakBackend('fr-fr')
out = backend._phonemize_aux(text, separator.Separator(), True)
assert out == [
'ʒɛm lɑ̃ɡlɛ', 'ʒɛm lə- (en)fʊtbɔːl(fr)', '(en)fʊtbɔːl(fr)',
'syʁtu lə- (en)ɹiəl(fr) madʁid', 'nytiliz pa (en)ɡuːɡəl(fr)'
]
backend = EspeakBackend('fr-fr', language_switch='remove-flags')
out = backend._phonemize_aux(text, separator.Separator(), True)
assert out == [
'ʒɛm lɑ̃ɡlɛ', 'ʒɛm lə- fʊtbɔːl', 'fʊtbɔːl', 'syʁtu lə- ɹiəl madʁid',
'nytiliz pa ɡuːɡəl'
]
backend = EspeakBackend('fr-fr', language_switch='remove-utterance')
out = backend._phonemize_aux(text, separator.Separator(), True)
assert out == ['ʒɛm lɑ̃ɡlɛ']
with pytest.raises(RuntimeError):
backend = EspeakBackend('fr-fr', language_switch='foo')
def test_french():
backend = EspeakBackend('fr-fr')
text = u'bonjour le monde'
sep = separator.Separator(word=';eword ', syllable=None, phone=' ')
expected = [u'b ɔ̃ ʒ u ʁ ;eword l ə ;eword m ɔ̃ d ;eword ']
out = backend._phonemize_aux(text, sep, False)
assert out == expected
def test_separator_3():
backend = SegmentsBackend('cree')
text = 'achi acho'
sep = separator.Separator(word=' ', syllable=None, phone='_')
assert backend.phonemize(text, separator=sep) == u'ʌ_tʃ_ɪ_ ʌ_tʃ_ʊ_ '
assert backend.phonemize(text, separator=sep, strip=True) \
== u'ʌ_tʃ_ɪ ʌ_tʃ_ʊ'
def text2phone(text, char2code):
seperator = separator.Separator('', '', ' ')
ph = phonemize.phonemize(text, separator=seperator)
ph = ph.split(' ')
ph.remove('')
result = [char2code[p] for p in ph]
return torch.LongTensor(result)
def setup(self):
# just a name shortcut
self.p = lambda text: phonemize(text,
language='en-us',
backend='festival',
strip=True,
separator=separator.Separator(
' ', '|', '-'))
def test_separator_5():
backend = SegmentsBackend('cree')
text = 'achi acho'
sep = separator.Separator(phone=' ', word='_')
assert backend.phonemize(text, separator=sep) == u'ʌ tʃ ɪ _ʌ tʃ ʊ _'
assert backend.phonemize(text, separator=sep, strip=True) \
== u'ʌ tʃ ɪ_ʌ tʃ ʊ'
def main():
"""Phonemize a text from command-line arguments"""
args = parse_args()
if args.version:
print(version())
return
# configure logging according to --verbose option. If verbose,
# init a logger to output on stderr. Else init a logger going to
# the void.
logger = logging.getLogger()
logger.handlers = []
logger.setLevel(logging.DEBUG)
if args.verbose:
handler = logging.StreamHandler(sys.stderr)
handler.setFormatter(logging.Formatter('%(message)s'))
else:
handler = logging.NullHandler()
logger.addHandler(handler)
# configure input as a readable stream
streamin = args.input
if isinstance(streamin, str):
streamin = codecs.open(streamin, 'r', encoding='utf8')
logger.debug('reading from %s', streamin.name)
# configure output as a writable stream
streamout = args.output
if isinstance(streamout, str):
streamout = codecs.open(streamout, 'w', 'utf8')
logger.debug('writing to %s', streamout.name)
# configure the separator for phonemes, syllables and words.
sep = separator.Separator(phone=args.phone_separator,
syllable=args.syllable_separator,
word=args.word_separator)
logger.debug('separator is %s', sep)
# load the input text (python2 optionnally needs an extra decode)
text = streamin.read()
try:
text = text.decode('utf8')
except (AttributeError, UnicodeEncodeError):
pass
# phonemize the input text
out = phonemize.phonemize(text,
language=args.language,
backend=args.backend,
separator=sep,
strip=args.strip,
use_sampa=args.sampa,
njobs=args.njobs,
logger=logger)
if len(out):
streamout.write(out + '\n')
def test_separator_4():
backend = SegmentsBackend('cree')
text = 'achi acho'
# TODO bug when sep.phone == ' ' with no sep.word
sep = separator.Separator(phone=' ', word='')
assert backend.phonemize(text, separator=sep) == u'ʌ tʃ ɪ ʌ tʃ ʊ '
assert backend.phonemize(text, separator=sep, strip=True) \
== u'ʌ tʃ ɪʌ tʃ ʊ'
def test_punctuation(text, strip, sep):
if sep == separator.Separator():
expected = 'ɐ kɑːmə ɐ pɔɪnt' if strip else 'ɐ kɑːmə ɐ pɔɪnt '
else:
expected = (
'ɐ_k ɑː m ə_ɐ_p ɔɪ n t' if strip else 'ɐ _k ɑː m ə _ɐ _p ɔɪ n t _')
output = EspeakBackend('en-us').phonemize(text, strip=strip, separator=sep)
assert expected == output
def text2phone(text, char2code):
seperator = separator.Separator('', '', ' ')
ph = phonemize(text, backend="festival", separator=seperator)
ph = ph.split(' ')
ph.remove('')
print('text %s ~ ph %s' % (text, ph))
result = [char2code[p] for p in ph]
return torch.LongTensor(result)
def main():
"""Phonemize a text from command-line arguments"""
args = parse_args()
if args.version:
print(version.version())
return
# configure logging according to --verbose/--quiet options
verbosity = 'normal'
if args.verbose:
verbosity = 'verbose'
elif args.quiet:
verbosity = 'quiet'
log = logger.get_logger(verbosity=verbosity)
# configure input as a readable stream
streamin = args.input
if isinstance(streamin, str):
streamin = codecs.open(streamin, 'r', encoding='utf8')
log.debug('reading from %s', streamin.name)
# configure output as a writable stream
streamout = args.output
if isinstance(streamout, str):
streamout = codecs.open(streamout, 'w', 'utf8')
log.debug('writing to %s', streamout.name)
# configure the separator for phonemes, syllables and words.
sep = separator.Separator(phone=args.phone_separator,
syllable=args.syllable_separator,
word=args.word_separator)
log.debug('separator is %s', sep)
# load the input text (python2 optionnally needs an extra decode)
text = streamin.read()
try:
text = text.decode('utf8')
except (AttributeError, UnicodeEncodeError):
pass
# phonemize the input text
out = phonemize.phonemize(text,
language=args.language,
backend=args.backend,
separator=sep,
strip=args.strip,
with_stress=args.with_stress,
use_sampa=args.sampa,
language_switch=args.language_switch,
njobs=args.njobs,
logger=log)
if len(out):
streamout.write(out + '\n')
def test_arabic():
backend = EspeakBackend('ar')
text = u'السلام عليكم'
sep = separator.Separator()
# Arabic seems to have changed starting at espeak-ng-1.49.3
if tuple(EspeakBackend.version().split('.')) >= ('1', '49', '3'):
expected = [u'ʔassalaːm ʕliːkm ']
else:
expected = [u'ʔassalaam ʕaliijkum ']
out = backend._phonemize_aux(text, sep, False)
assert out == expected
def test_phone_separator_simple():
text = 'The lion and the tiger ran'
sep = separator.Separator(phone='_')
backend = EspeakBackend('en-us')
output = backend.phonemize(text, separator=sep, strip=True)
expected = 'ð_ə l_aɪə_n æ_n_d ð_ə t_aɪ_ɡ_ɚ ɹ_æ_n'
assert expected == output
output = backend.phonemize(text, separator=sep, strip=False)
expected = 'ð_ə_ l_aɪə_n_ æ_n_d_ ð_ə_ t_aɪ_ɡ_ɚ_ ɹ_æ_n_ '
assert expected == output
def test_french_sampa():
text = u'bonjour le monde'
backend = EspeakMbrolaBackend('mb-fr1')
sep = separator.Separator(word=None, phone=' ')
expected = 'b o~ Z u R l @ m o~ d '
out = backend.phonemize(text, separator=sep, strip=False)
assert out == expected
expected = 'b o~ Z u R l @ m o~ d'
out = backend.phonemize(text, separator=sep, strip=True)
assert out == expected
assert '' == backend.phonemize('', separator=sep, strip=True)
assert '' == backend.phonemize('"', separator=sep, strip=True)
def main():
"""Phonemize a text from command-line arguments"""
args = parse_args()
# setup a custom path to espeak and festival if required (this must be done
# before generating the version message)
if args.espeak_path:
EspeakBackend.set_espeak_path(args.espeak_path)
if args.festival_path:
FestivalBackend.set_festival_path(args.festival_path)
# display version information and exit
if args.version:
print(version.version())
return
# list supported languages and exit
if args.list_languages:
backends = (['festival', 'segments', 'espeak', 'espeak-mbrola']
if not args.backend else [args.backend])
for backend in backends:
print(f'supported languages for {backend} are:\n' +
'\n'.join(f'\t{k}\t->\t{v}' for k, v in sorted(
BACKENDS_MAP[backend].supported_languages().items())))
return
# set default backend as espeak if not specified
args.backend = args.backend or 'espeak'
# configure logging according to --verbose/--quiet options
verbosity = 'normal'
if args.verbose:
verbosity = 'verbose'
elif args.quiet:
verbosity = 'quiet'
log = logger.get_logger(verbosity=verbosity)
# configure input as a readable stream
streamin = args.input
if isinstance(streamin, str):
streamin = codecs.open(streamin, 'r', encoding='utf8')
log.debug('reading from %s', streamin.name)
# configure output as a writable stream
streamout = args.output
if isinstance(streamout, str):
streamout = codecs.open(streamout, 'w', 'utf8')
log.debug('writing to %s', streamout.name)
# configure the separator for phonemes, syllables and words.
if args.backend == 'espeak-mbrola':
log.debug('using espeak-mbrola backend: ignoring word separator')
sep = separator.Separator(phone=args.phone_separator,
syllable=None,
word=None)
else:
sep = separator.Separator(phone=args.phone_separator,
syllable=args.syllable_separator,
word=args.word_separator)
log.debug('separator is %s', sep)
text = [line.strip() for line in streamin]
# phonemize the input text
out = phonemize(text,
language=args.language,
backend=args.backend,
separator=sep,
strip=args.strip,
preserve_punctuation=args.preserve_punctuation,
punctuation_marks=args.punctuation_marks,
with_stress=args.with_stress,
language_switch=args.language_switch,
njobs=args.njobs,
logger=log)
if out:
streamout.write('\n'.join(out) + '\n')
def _test(text, separator=separator.Separator(' ', '|', '-')):
return festival.phonemize(text,
language='en-us',
strip=True,
separator=separator)
def test_phone_separator(text, expected):
sep = separator.Separator(phone='_')
backend = EspeakBackend('en-us')
output = backend.phonemize(text, separator=sep, strip=True)
assert output == expected
def test_im():
sep = separator.Separator(word=' ', syllable='', phone='')
assert _test("I'm looking for an image", sep) \
== ['aym luhkaxng faor axn ihmaxjh']
assert _test("Im looking for an image", sep) \
== ['ihm luhkaxng faor axn ihmaxjh']
def collate_fn_common(batch,
data_type,
max_len_mel=2000,
reconstructed_phoneme=False):
#
ggg, truefalse = batch[0]
batch_size = len(batch)
parts = 6
final_list = []
if truefalse:
seq_len = len(ggg[1])
del ggg
example_id = []
for n in range(parts):
final_list.append([None] * (batch_size * seq_len))
#
for i in range(batch_size):
part, truefalse = batch[i]
for j in range(parts):
final_list[j][i * seq_len:(i + 1) * seq_len] = part[j]
example_id.append(i * seq_len)
#
else:
del ggg
for n in range(parts):
final_list.append([None] * (batch_size))
#
for i in range(batch_size):
part, truefalse = batch[i]
waveform, sample_rate, client_id, sentence = part
part = [
waveform, waveform.shape[1], sample_rate, client_id, sentence,
len(sentence)
]
for j in range(parts):
final_list[j][i] = part[j]
example_id = None
#
waveforms, waveform_l, sample_rates, client_ids, sentences, sentences_l = final_list
#
if not reconstructed_phoneme:
sentences = phonemize(sentences,
backend='espeak',
with_stress=False,
separator=separator.Separator(phone=' ',
syllable='',
word='- '))
for i in range(len(sentences_l)):
sentences_l[i] = len(sentences[i])
biggest_l_index = sentences_l.index(max(sentences_l))
token = tokenizer.tokenize(sentences[biggest_l_index])
text_field = TextField(token, token_indexer)
text_field.index(vocab)
padding_lengths = text_field.get_padding_lengths()
list_tokens = []
mel_list = [None] * len(sample_rates)
mel_list_l = [None] * len(sample_rates)
#
for i in range(len(sentences_l)):
token = tokenizer.tokenize(sentences[i])
text_field = TextField(token, token_indexer)
text_field.index(vocab)
tensor_dict = text_field.as_tensor(padding_lengths)
list_tokens.append(tensor_dict)
#
if data_type == "train":
mel_list[i] = train_audio_transforms(
waveforms[i]).squeeze(0).transpose(0, 1)
else:
mel_list[i] = test_audio_transforms(
waveforms[i]).squeeze(0).transpose(0, 1)
mel_list_l[i] = mel_list[i].shape[0]
waveforms[i] = waveforms[i].squeeze(0)
#
waveforms = nn.utils.rnn.pad_sequence(waveforms,
batch_first=True).unsqueeze(1)
#
mel_list.append(torch.zeros((max_len_mel, mel_list[0].shape[1])))
spectrograms = nn.utils.rnn.pad_sequence(
mel_list, batch_first=True).unsqueeze(1).transpose(2, 3)
spectrograms = spectrograms[1:]
highest_mel_l = spectrograms[0].shape[2]
mel_mask = create_mask_pad(highest_mel_l, mel_list_l)
#
text_field_tensors = text_field.batch_tensors(list_tokens)
#
sentences_tensor = nn_util.get_token_ids_from_text_field_tensors(
text_field_tensors)
sentences_mask = nn_util.get_text_field_mask(text_field_tensors) == False
return sentences_tensor, sentences_mask, spectrograms, mel_mask, waveforms, waveform_l, client_ids, example_id
def test_im():
sep = separator.Separator(' ', '', '')
assert _test("I'm looking for an image", sep) \
== ['aym luhkaxng faor axn ihmaxjh']
assert _test("Im looking for an image", sep) \
== ['ihm luhkaxng faor axn ihmaxjh']
def recognize(self, wav=None):
if not self.eng.find("fest") == -1:
# ======= festival english us only =====================================
from phonemizer.backend import FestivalBackend
out1 = FestivalBackend(
'en-us', preserve_punctuation=False).phonemize(self.text,
strip=True)
elif not self.eng.find("esp") == -1:
# ======= espeak =======================================================
from phonemizer.backend import EspeakBackend
if not self.lan.find("en") == -1:
backend = EspeakBackend('en-us')
elif not self.lan.find("fr") == -1:
backend = EspeakBackend('fr-fr')
elif not self.lan.find("de") == -1 or not self.lan.find(
"ger") == -1:
backend = EspeakBackend('de-de')
elif not self.lan.find("ita") == -1:
backend = EspeakBackend('it-it')
elif not self.lan.find("esp") == -1 or not self.lan.find(
"spa") == -1:
backend = EspeakBackend('es-es')
else:
print("Invalid Language specified")
exit(1)
sep = separator.Separator(word=';eword ', syllable=None, phone=' ')
out1 = backend.phonemize(self.text, sep, False)
elif not self.eng.find("mbr") == -1:
# ====== mbrola ========================================================
from phonemizer.backend import EspeakBackend, EspeakMbrolaBackend
if EspeakMbrolaBackend.is_available():
if not self.lan.find("en") == -1:
EspeakMbrolaBackend('mb-en1')
elif not self.lan.find("fr") == -1:
EspeakMbrolaBackend('mb-fr2')
elif not self.lan.find("de") == -1 or not self.lan.find(
"ger") == -1:
EspeakMbrolaBackend('mb-de1')
elif not self.lan.find("ita") == -1:
EspeakMbrolaBackend('mb-es1')
elif not self.lan.find("esp") == -1 or not self.lan.find(
"spa") == -1:
EspeakMbrolaBackend('mb-it3')
else:
print("Invalid Language specified")
exit(1)
sep = separator.Separator(word=';eword ',
syllable=None,
phone=' ')
out1 = backend.phonemize(self.text, sep, False)
elif not self.eng.find("seg") == -1:
# ===== segment ========================================================
from phonemizer.backend import SegmentsBackend
if not self.lan.find("ja") == -1:
backend = SegmentsBackend('japanese')
else:
print("Invalid Language specified")
exit(1)
sep = separator.Separator(word=';eword ', syllable=None, phone=' ')
out1 = backend.phonemize(self.text, sep, False)
else:
print("Invalid Language specified")
exit(1)
print(out1) # return the result string to the speach engine
def test_french():
text = u'bonjour le monde'
sep = separator.Separator(word=';eword ', syllable=None, phone=' ')
expected = [u'b ɔ̃ ʒ u ʁ ;eword l ə- ;eword m ɔ̃ d ;eword ']
out = espeak.phonemize(text, language='fr-fr', separator=sep, strip=False)
assert out == expected
'ʒɛm lɑ̃ɡlɛ', 'ʒɛm lə fʊtbɔːl', 'fʊtbɔːl', 'syʁtu lə ɹiəl madʁid',
'nytiliz pa ɡuːɡəl'
]
backend = EspeakBackend('fr-fr', language_switch='remove-utterance')
out = backend.phonemize(text, separator.Separator(), True)
assert out == ['ʒɛm lɑ̃ɡlɛ']
with pytest.raises(RuntimeError):
backend = EspeakBackend('fr-fr', language_switch='foo')
@pytest.mark.parametrize('text, strip, sep', ((t, s, u) for t in [
'a comma a point', 'a comma. a point.', 'a comma,, a point.',
'a comma, , a point.', 'a comma? a point!'
] for s in (True, False) for u in (separator.Separator(),
separator.Separator(word='_', phone=' '))))
def test_punctuation(text, strip, sep):
if sep == separator.Separator():
expected = 'ɐ kɑːmə ɐ pɔɪnt' if strip else 'ɐ kɑːmə ɐ pɔɪnt '
else:
expected = ('ɐ_k ɑː m ə_ɐ_p ɔɪ n t'
if strip else 'ɐ _k ɑː m ə _ɐ _p ɔɪ n t _')
output = EspeakBackend('en-us').phonemize(text, strip=strip, separator=sep)
assert expected == output
# see https://github.com/bootphon/phonemizer/issues/31
def test_phone_separator_simple():
text = 'The lion and the tiger ran'
def main():
"""Phonemize a text from command-line arguments"""
args = parse_args()
# setup a custom path to espeak and festival if required (this must be done
# before generating the version message)
if args.espeak_library:
BACKENDS['espeak'].set_library(args.espeak_library)
if args.festival_executable:
BACKENDS['festival'].set_executable(args.festival_executable)
# display version information and exit
if args.version:
print(version.version())
return
# list supported languages and exit
if args.list_languages:
print(list_languages(args.backend))
return
# set default backend as espeak if not specified
args.backend = args.backend or 'espeak'
# configure logging according to --verbose/--quiet options
log = get_logger(args.verbose, args.quiet)
# configure input:output as a readable/writable streams
streamin = setup_stream(args.input, 'r')
log.debug('reading from %s', streamin.name)
streamout = setup_stream(args.output, 'w')
log.debug('writing to %s', streamout.name)
# configure the separator for phonemes, syllables and words.
if args.backend == 'espeak-mbrola':
log.debug('using espeak-mbrola backend: ignoring word separator')
sep = separator.Separator(phone=args.phone_separator,
syllable=None,
word=None)
else:
sep = separator.Separator(phone=args.phone_separator,
syllable=args.syllable_separator,
word=args.word_separator)
log.debug('separator is %s', sep)
if args.prepend_text:
input_output_separator = sep.input_output_separator(args.prepend_text)
log.debug('prepend input text to output, separator is "%s"',
input_output_separator)
else:
input_output_separator = False
# phonemize the input text
out = phonemize(streamin.readlines(),
language=args.language,
backend=args.backend,
separator=sep,
strip=args.strip,
prepend_text=args.prepend_text,
preserve_punctuation=args.preserve_punctuation,
punctuation_marks=args.punctuation_marks,
with_stress=args.with_stress,
tie=args.tie,
language_switch=args.language_switch,
words_mismatch=args.words_mismatch,
njobs=args.njobs,
logger=log)
if out and input_output_separator:
streamout.write(
os.linesep.join(f'{line[0]} {input_output_separator} {line[1]}'
for line in out) + os.linesep)
elif out:
streamout.write(os.linesep.join(out) + os.linesep)
assert out == ['ʒɛm lɑ̃ɡlɛ']
with pytest.raises(RuntimeError):
backend = EspeakBackend('fr-fr', language_switch='foo')
@pytest.mark.parametrize(
'text, strip, sep',
((t, s, u) for t in [
'a comma a point',
'a comma. a point.',
'a comma,, a point.',
'a comma, , a point.',
'a comma? a point!']
for s in (True, False)
for u in (separator.Separator(), separator.Separator(word='_', phone=' '))
))
def test_punctuation(text, strip, sep):
if sep == separator.Separator():
expected = 'ɐ kɑːmə ɐ pɔɪnt' if strip else 'ɐ kɑːmə ɐ pɔɪnt '
else:
expected = (
'ɐ_k ɑː m ə_ɐ_p ɔɪ n t' if strip else 'ɐ _k ɑː m ə _ɐ _p ɔɪ n t _')
output = EspeakBackend('en-us').phonemize(text, strip=strip, separator=sep)
assert expected == output
# see https://github.com/bootphon/phonemizer/issues/31
def test_phone_separator_simple():
text = 'The lion and the tiger ran'
def _test(text):
return phonemize(text,
language='en-us',
backend='festival',
strip=True,
separator=separator.Separator(' ', '|', '-'))
def _test(text, separator=separator.Separator(
word=' ', syllable='|', phone='-')):
backend = FestivalBackend('en-us')
return backend._phonemize_aux(text, separator, True)
