Generates Cued Speech keys for given tokens and their pronunciations.
Unlike :class:sppasWhatKeyPredictor, which works on time-aligned phoneme
tiers, this class works directly on pronunciation strings (e.g. from a
pronunciation dictionary), without requiring any time alignment.
Create a new instance.
def __init__(self, cued_rules: CuedSpeechKeys):
"""Create a new instance.
:param cued_rules: (CuedSpeechKeys) The rules of the cued speech.
"""
if isinstance(cued_rules, CuedSpeechKeys) is False:
raise TypeError(f'Given cued_rules must be a CuedSpeechKeys object. Got {type(cued_rules)} instead.')
self.__cs = cued_rules
self.__cue_tokenizer = CueingPronTokenizer(self.__cs)
self.__keys_by_token = CueingKeysByToken()
self._stops = list(symbols.phone.keys())
self._stops.append('#')
self._stops.append('*')
self._stops.append('@@')
self._stops.append('+')
self._stops.append('sil')
self._stops.append('sp')
self._stops.append('gb')
self._stops.append('lg')
self._stops.append('fp')
self._stops.append('dummy')
self._stops.append('noise')
self._stops.append('laugh')
Return the sequence of keys from the list of tokens and their pronunciations.
Input tokens: ["test", "hello"] Input prons: ['t-E-s-t', 'E-l-o'] Code result: [('5-c.3-s.5-c.6-s', 't-E.s-vnil.t-E.l-o')] Returned keys by token: (('5-c.3-s', '5-c.6-s'), ('t-E.s-vnil', 't-E.l-o'))
def annotate(self, tokens: list, prons: list) -> tuple:
"""Return the sequence of keys from the list of tokens and their pronunciations.
Input tokens: ["test", "hello"]
Input prons: ['t-E-s-t', 'E-l-o']
Code result: [('5-c.3-s.5-c.6-s', 't-E.s-vnil.t-E.l-o')]
Returned keys by token: (('5-c.3-s', '5-c.6-s'), ('t-E.s-vnil', 't-E.l-o'))
:param tokens: (list) List of tokens
:param prons: (list) List of pronunciations
:raises: TypeError: Invalid text type.
:return: tuple(str) List of keys
"""
if type(prons) not in (list, tuple):
raise TypeError(f"Given pronunciations must be a list. Got '{type(prons)}' instead.")
if type(tokens) not in (list, tuple):
raise TypeError(f"Given tokens must be a list. Got '{type(tokens)}' instead.")
cued = self._cuer(prons)
normalized_word_phonemes = self.__cue_tokenizer.normalize_word_phonemes(tuple(prons))
(codes, phons) = self.__keys_by_token.segment(normalized_word_phonemes, cued)
return (codes, phons)
Return the result of "CuedSpeech" on the given phonetized text.
def _cuer(self, token_prons: list) -> list:
"""Return the result of "CuedSpeech" on the given phonetized text.
:param token_prons: (list) Pronunciation of each token.
:return: (list) List of keys
"""
results = list()
prons = list()
for token_pron in token_prons:
prons.extend(token_pron.split('-'))
phonemes = list()
for p in prons:
if p in self._stops:
if len(phonemes) > 0:
sgmts = self.__cs.syllabify(phonemes)
phons = self.__cs.phonetize_syllables(phonemes, sgmts)
keys = self.__cs.keys_phonetized(phons)
results.append((keys, phons))
phonemes = list()
if p in ('#', 'sil', 'lg', 'laugh'):
results.append(('0-n', 'cnil-' + p))
else:
results.append(('0-s', 'cnil-' + p))
else:
phonemes.append(p)
if len(phonemes) > 0:
sgmts = self.__cs.syllabify(phonemes)
phons = self.__cs.phonetize_syllables(phonemes, sgmts)
keys = self.__cs.keys_phonetized(phons)
results.append((keys, phons))
return results