def update_symbols(data: MergedDataset, symbols: SymbolIdDict) -> SymbolIdDict:
new_symbols: Set[str] = {
x
for y in data.items()
for x in symbols.get_symbols(y.serialized_symbol_ids)
}
new_symbol_ids = SymbolIdDict.init_from_symbols_with_pad(
new_symbols, pad_symbol=DEFAULT_PADDING_SYMBOL)
if new_symbol_ids.get_all_symbols() != symbols.get_all_symbols():
for entry in data.items():
original_symbols = symbols.get_symbols(entry.serialized_symbol_ids)
entry.serialized_symbol_ids = new_symbol_ids.get_serialized_ids(
original_symbols)
return new_symbol_ids
def sents_convert_to_ipa(sentences: SentenceList, text_symbols: SymbolIdDict, ignore_tones: bool, ignore_arcs: bool, mode: Optional[EngToIpaMode], consider_ipa_annotations: bool, logger: Logger) -> Tuple[SymbolIdDict, SentenceList]:
sents_new_symbols = []
for sentence in sentences.items(True):
if sentence.lang == Language.ENG and mode is None:
ex = "Please specify the ipa conversion mode."
logger.exception(ex)
raise Exception(ex)
new_symbols, new_accent_ids = symbols_to_ipa(
symbols=text_symbols.get_symbols(sentence.serialized_symbols),
lang=sentence.lang,
accent_ids=deserialize_list(sentence.serialized_accents),
ignore_arcs=ignore_arcs,
ignore_tones=ignore_tones,
mode=mode,
replace_unknown_with=DEFAULT_PADDING_SYMBOL,
consider_ipa_annotations=consider_ipa_annotations,
logger=logger,
)
assert len(new_symbols) == len(new_accent_ids)
sentence.lang = Language.IPA
sentence.serialized_accents = serialize_list(new_accent_ids)
sents_new_symbols.append(new_symbols)
assert len(sentence.get_accent_ids()) == len(new_symbols)
return update_symbols_and_text(sentences, sents_new_symbols)
def get_formatted(self, symbol_id_dict: SymbolIdDict, accent_id_dict: AccentsDict, pairs_per_line=170, space_length=0):
return get_formatted_core(
sent_id=self.sent_id,
symbols=symbol_id_dict.get_symbols(self.serialized_symbols),
accent_ids=self.get_accent_ids(),
accent_id_dict=accent_id_dict,
space_length=space_length,
max_pairs_per_line=pairs_per_line
)
def sents_map(sentences: SentenceList, text_symbols: SymbolIdDict, symbols_map: SymbolsMap, ignore_arcs: bool, logger: Logger) -> Tuple[SymbolIdDict, SentenceList]:
sents_new_symbols = []
result = SentenceList()
new_sent_id = 0
ipa_settings = IPAExtractionSettings(
ignore_tones=False,
ignore_arcs=ignore_arcs,
replace_unknown_ipa_by=DEFAULT_PADDING_SYMBOL,
)
for sentence in sentences.items():
symbols = text_symbols.get_symbols(sentence.serialized_symbols)
accent_ids = deserialize_list(sentence.serialized_accents)
mapped_symbols = symbols_map.apply_to_symbols(symbols)
text = SymbolIdDict.symbols_to_text(mapped_symbols)
# a resulting empty text would make no problems
sents = text_to_sentences(
text=text,
lang=sentence.lang,
logger=logger,
)
for new_sent_text in sents:
new_symbols = text_to_symbols(
new_sent_text,
lang=sentence.lang,
ipa_settings=ipa_settings,
logger=logger,
)
if len(accent_ids) > 0:
new_accent_ids = [accent_ids[0]] * len(new_symbols)
else:
new_accent_ids = []
assert len(new_accent_ids) == len(new_symbols)
new_sent_id += 1
tmp = Sentence(
sent_id=new_sent_id,
text=new_sent_text,
lang=sentence.lang,
orig_lang=sentence.orig_lang,
# this is not correct but nearest possible currently
original_text=sentence.original_text,
serialized_accents=serialize_list(new_accent_ids),
serialized_symbols=""
)
sents_new_symbols.append(new_symbols)
assert len(tmp.get_accent_ids()) == len(new_symbols)
result.append(tmp)
return update_symbols_and_text(result, sents_new_symbols)
def get_ngram_rarity(data: PreparedDataList, corpus: PreparedDataList,
symbols: SymbolIdDict,
ngram: int) -> OrderedDictType[int, float]:
data_symbols_dict = OrderedDict({
x.entry_id: symbols.get_symbols(x.serialized_symbol_ids)
for x in data.items()
})
corpus_symbols_dict = OrderedDict({
x.entry_id: symbols.get_symbols(x.serialized_symbol_ids)
for x in corpus.items()
})
rarity = get_rarity_ngrams(
data=data_symbols_dict,
corpus=corpus_symbols_dict,
n_gram=ngram,
ignore_symbols=None,
)
return rarity
def from_sentences(cls, sentences: SentenceList, accents: AccentsDict, symbols: SymbolIdDict):
res = cls()
for sentence in sentences.items():
infer_sent = InferSentence(
sent_id=sentence.sent_id,
symbols=symbols.get_symbols(sentence.serialized_symbols),
accents=accents.get_accents(sentence.serialized_accents),
original_text=sentence.original_text,
)
assert len(infer_sent.symbols) == len(infer_sent.accents)
res.append(infer_sent)
return res
def sents_accent_template(sentences: SentenceList, text_symbols: SymbolIdDict, accent_ids: AccentsDict) -> AccentedSymbolList:
res = AccentedSymbolList()
for i, sent in enumerate(sentences.items()):
symbols = text_symbols.get_symbols(sent.serialized_symbols)
accents = accent_ids.get_accents(sent.serialized_accents)
for j, symbol_accent in enumerate(zip(symbols, accents)):
symbol, accent = symbol_accent
accented_symbol = AccentedSymbol(
position=f"{i}-{j}",
symbol=symbol,
accent=accent
)
res.append(accented_symbol)
return res
def sents_normalize(sentences: SentenceList, text_symbols: SymbolIdDict, logger: Logger) -> Tuple[SymbolIdDict, SentenceList]:
# Maybe add info if something was unknown
sents_new_symbols = []
for sentence in sentences.items():
new_symbols, new_accent_ids = symbols_normalize(
symbols=text_symbols.get_symbols(sentence.serialized_symbols),
lang=sentence.lang,
accent_ids=deserialize_list(sentence.serialized_accents),
logger=logger,
)
# TODO: check if new sentences resulted and then split them.
sentence.serialized_accents = serialize_list(new_accent_ids)
sents_new_symbols.append(new_symbols)
return update_symbols_and_text(sentences, sents_new_symbols)
def get_ngram_stats_df(symbols: SymbolIdDict, trainset: PreparedDataList,
valset: PreparedDataList, testset: PreparedDataList,
restset: PreparedDataList, n: int, logger: Logger):
total_set = get_total_set(trainset, valset, testset, restset)
logger.info(f"Getting all {n}-gram stats...")
tot_symbols = [
symbols.get_symbols(x.serialized_symbol_ids)
for x in total_set.items()
]
tot_symbols_one_gram = [get_ngrams(x, n=n) for x in tot_symbols]
symbol_order = list(sorted({x for y in tot_symbols_one_gram for x in y}))
ngram_stats = _get_ngram_stats_df_core(
symbol_order=symbol_order,
symbols=symbols,
trainset=trainset,
valset=valset,
testset=testset,
restset=restset,
n=n,
logger=logger,
)
occurences_count_df, occurrences_percent_df, occurrences_distribution_percent_df, utterance_occurrences_count_df, utterance_occurrences_percent_df, uniform_occurrences_count_df, uniform_occurrences_percent_df = ngram_stats
symbol_dfs = []
symbol_dfs.append(occurences_count_df)
symbol_dfs.append(occurrences_percent_df)
symbol_dfs.append(occurrences_distribution_percent_df)
symbol_dfs.append(utterance_occurrences_count_df)
symbol_dfs.append(utterance_occurrences_percent_df)
symbol_dfs.append(uniform_occurrences_count_df)
symbol_dfs.append(uniform_occurrences_percent_df)
for i in range(1, len(symbol_dfs)):
symbol_dfs[i] = symbol_dfs[i].loc[:, symbol_dfs[i].
columns != FIRST_COL_NAME]
symbol_stats = pd.concat(
symbol_dfs,
axis=1,
join='inner',
)
# symbol_stats = symbol_stats.round(decimals=2)
symbol_stats = symbol_stats.sort_values(by='TOTAL_OCCURRENCES_COUNT',
ascending=False)
print(symbol_stats)
return symbol_stats
def log_stats(data: MergedDataset, symbols: SymbolIdDict,
accent_ids: AccentsDict, speakers: SpeakersDict, logger: Logger):
logger.info(
f"Speakers ({len(speakers)}): {', '.join(sorted(speakers.get_all_speakers()))}"
)
logger.info(
f"Symbols ({len(symbols)}): {' '.join(sorted(symbols.get_all_symbols()))}"
)
logger.info(
f"Accents ({len(accent_ids)}): {', '.join(sorted(accent_ids.get_all_accents()))}"
)
logger.info(
f"Entries ({len(data)}): {data.get_total_duration_s()/60:.2f}m")
symbol_counter = get_counter(
[symbols.get_symbols(x.serialized_symbol_ids) for x in data.items()])
logger.info(symbol_counter)
def prep_data_list_to_dict_with_symbols(l: PreparedDataList, symbols: SymbolIdDict) -> OrderedDictType[int, List[str]]:
res = OrderedDict({x.entry_id: symbols.get_symbols(x.serialized_symbol_ids) for x in l.items()})
return res
def get_formatted_v2(self, symbol_id_dict: SymbolIdDict):
return get_formatted_core_v2(
sent_id=self.sent_id,
symbols=symbol_id_dict.get_symbols(self.serialized_symbols),
original_text=self.original_text,
)
def _get_ngram_stats_df_core(symbol_order: List[str], symbols: SymbolIdDict,
trainset: PreparedDataList,
valset: PreparedDataList,
testset: PreparedDataList,
restset: PreparedDataList, n: int,
logger: Logger):
logger.info(f"Get {n}-grams...")
trn_symbols = [
symbols.get_symbols(x.serialized_symbol_ids) for x in trainset.items()
]
val_symbols = [
symbols.get_symbols(x.serialized_symbol_ids) for x in valset.items()
]
tst_symbols = [
symbols.get_symbols(x.serialized_symbol_ids) for x in testset.items()
]
rst_symbols = [
symbols.get_symbols(x.serialized_symbol_ids) for x in restset.items()
]
trn_symbols_one_gram = [get_ngrams(x, n=n) for x in trn_symbols]
val_symbols_one_gram = [get_ngrams(x, n=n) for x in val_symbols]
tst_symbols_one_gram = [get_ngrams(x, n=n) for x in tst_symbols]
rst_symbols_one_gram = [get_ngrams(x, n=n) for x in rst_symbols]
logger.info("Get stats...")
occurences_count_df = get_occ_df_of_all_symbols(
symbols=symbol_order,
data_trn=trn_symbols_one_gram,
data_val=val_symbols_one_gram,
data_tst=tst_symbols_one_gram,
data_rst=rst_symbols_one_gram,
)
occurences_count_df.columns = [
FIRST_COL_NAME, 'TRAIN_OCCURRENCES_COUNT', 'VAL_OCCURRENCES_COUNT',
'TEST_OCCURRENCES_COUNT', 'REST_OCCURRENCES_COUNT',
'TOTAL_OCCURRENCES_COUNT'
]
print(occurences_count_df)
occurrences_percent_df = get_rel_occ_df_of_all_symbols(occurences_count_df)
occurrences_percent_df.columns = [
FIRST_COL_NAME, 'TRAIN_OCCURRENCES_PERCENT', 'VAL_OCCURRENCES_PERCENT',
'TEST_OCCURRENCES_PERCENT', 'REST_OCCURRENCES_PERCENT'
]
print(occurrences_percent_df)
occurrences_distribution_percent_df = get_dist_among_other_symbols_df_of_all_symbols(
occs_df=occurences_count_df,
data_trn=trn_symbols_one_gram,
data_val=val_symbols_one_gram,
data_tst=tst_symbols_one_gram,
data_rst=rst_symbols_one_gram,
)
occurrences_distribution_percent_df.columns = [
FIRST_COL_NAME, 'TRAIN_OCCURRENCES_DISTRIBUTION_PERCENT',
'VAL_OCCURRENCES_DISTRIBUTION_PERCENT',
'TEST_OCCURRENCES_DISTRIBUTION_PERCENT',
'REST_OCCURRENCES_DISTRIBUTION_PERCENT',
'TOTAL_OCCURRENCES_DISTRIBUTION_PERCENT'
]
print(occurrences_distribution_percent_df)
utterance_occurrences_count_df = get_utter_occ_df_of_all_symbols(
symbols=symbol_order,
data_trn=trn_symbols_one_gram,
data_val=val_symbols_one_gram,
data_tst=tst_symbols_one_gram,
data_rst=rst_symbols_one_gram,
)
utterance_occurrences_count_df.columns = [
FIRST_COL_NAME, 'TRAIN_UTTERANCE_OCCURRENCES_COUNT',
'VAL_UTTERANCE_OCCURRENCES_COUNT', 'TEST_UTTERANCE_OCCURRENCES_COUNT',
'REST_UTTERANCE_OCCURRENCES_COUNT', 'TOTAL_UTTERANCE_OCCURRENCES_COUNT'
]
print(utterance_occurrences_count_df)
utterance_occurrences_percent_df = get_rel_utter_occ_df_of_all_symbols(
utterance_occurrences_count_df)
utterance_occurrences_percent_df.columns = [
FIRST_COL_NAME, 'TRAIN_UTTERANCE_OCCURRENCES_PERCENT',
'VAL_UTTERANCE_OCCURRENCES_PERCENT',
'TEST_UTTERANCE_OCCURRENCES_PERCENT',
'REST_UTTERANCE_OCCURRENCES_PERCENT'
]
print(utterance_occurrences_percent_df)
uniform_occurrences_count_df = get_uniform_distr_df_for_occs(
symbols=symbol_order,
occ_df=occurences_count_df,
)
uniform_occurrences_count_df.columns = [
FIRST_COL_NAME, 'TRAIN_UNIFORM_OCCURRENCES_COUNT',
'VAL_UNIFORM_OCCURRENCES_COUNT', 'TEST_UNIFORM_OCCURRENCES_COUNT',
'REST_UNIFORM_OCCURRENCES_COUNT', 'TOTAL_UNIFORM_OCCURRENCES_COUNT'
]
print(uniform_occurrences_count_df)
uniform_occurrences_percent_df = get_rel_uniform_distr_df_for_occs(
symbols=symbol_order, )
uniform_occurrences_percent_df.columns = [
FIRST_COL_NAME, 'UNIFORM_OCCURRENCES_PERCENT'
]
print(uniform_occurrences_percent_df)
return occurences_count_df, occurrences_percent_df, occurrences_distribution_percent_df, utterance_occurrences_count_df, utterance_occurrences_percent_df, uniform_occurrences_count_df, uniform_occurrences_percent_df
def int_set_to_symbols(symbol_ids: Optional[Set[int]],
symbols: SymbolIdDict) -> Optional[Set[str]]:
if symbol_ids is None:
return None
ignore_symbols = set(symbols.get_symbols(list(symbol_ids)))
return ignore_symbols
