def __init__(self, deterministic: bool = True):
super().__init__(name="ordinal", kind="verbalize", deterministic=deterministic)
graph_digit = pynini.string_file(get_abs_path("data/ordinals/digit.tsv")).invert()
graph_teens = pynini.string_file(get_abs_path("data/ordinals/teen.tsv")).invert()
graph = (
pynutil.delete("integer:")
+ delete_space
+ pynutil.delete("\"")
+ pynini.closure(NEMO_NOT_QUOTE, 1)
+ pynutil.delete("\"")
)
convert_rest = pynutil.insert("th", weight=0.01)
suffix = pynini.cdrewrite(
graph_digit | graph_teens | pynutil.add_weight(pynini.cross("ty", "tieth"), weight=0.001) | convert_rest,
"",
"[EOS]",
NEMO_SIGMA,
).optimize()
graph = graph @ suffix
self.suffix = suffix
delete_tokens = self.delete_tokens(graph)
self.fst = delete_tokens.optimize()
def __init__(self, deterministic: bool = True):
super().__init__(name="cardinal",
kind="classify",
deterministic=deterministic)
graph = pynini.Far(
get_abs_path("data/numbers/cardinal_number_name.far")).get_fst()
self.graph_hundred_component_at_least_one_none_zero_digit = (
pynini.closure(NEMO_DIGIT, 2, 3)
| pynini.difference(NEMO_DIGIT, pynini.accep("0"))) @ graph
self.graph = (pynini.closure(NEMO_DIGIT, 1, 3) + pynini.closure(
pynini.closure(pynutil.delete(","), 0, 1) + NEMO_DIGIT +
NEMO_DIGIT + NEMO_DIGIT)) @ graph
graph_digit = pynini.string_file(
get_abs_path("data/numbers/digit.tsv"))
graph_zero = pynini.string_file(get_abs_path("data/numbers/zero.tsv"))
single_digits_graph = pynutil.add_weight(
pynini.invert(graph_digit | graph_zero), 1.2) | pynutil.add_weight(
pynini.cross("0", "oh"), 1.1)
self.single_digits_graph = single_digits_graph + pynini.closure(
pynutil.insert(" ") + single_digits_graph)
if not deterministic:
single_digits_graph_with_commas = pynini.closure(
self.single_digits_graph + pynutil.insert(" "), 1,
3) + pynini.closure(
pynutil.delete(",") + single_digits_graph +
pynutil.insert(" ") + single_digits_graph +
pynutil.insert(" ") + single_digits_graph,
1,
)
self.graph |= self.single_digits_graph | get_hundreds_graph(
) | single_digits_graph_with_commas
self.range_graph = (
pynini.closure(pynutil.insert("from "), 0, 1) + self.graph +
(pynini.cross("-", " to ") | pynini.cross("-", " ")) +
self.graph)
self.range_graph |= self.graph + (pynini.cross(
"x", " by ") | pynini.cross(" x ", " by ")) + self.graph
self.range_graph = self.range_graph.optimize()
optional_minus_graph = pynini.closure(
pynutil.insert("negative: ") + pynini.cross("-", "\"true\" "), 0,
1)
final_graph = self.graph | pynutil.add_weight(self.get_serial_graph(),
1.2)
if not deterministic:
final_graph |= self.range_graph
final_graph = optional_minus_graph + pynutil.insert(
"integer: \"") + final_graph + pynutil.insert("\"")
final_graph = self.add_tokens(final_graph)
self.fst = final_graph.optimize()
def __init__(self, deterministic: bool = True):
super().__init__(name="electronic",
kind="verbalize",
deterministic=deterministic)
graph_digit_no_zero = pynini.invert(
pynini.string_file(
get_abs_path("data/numbers/digit.tsv"))).optimize()
graph_zero = pynini.cross("0", "zero")
if not deterministic:
graph_zero |= pynini.cross("0", "o") | pynini.cross("0", "oh")
graph_digit = graph_digit_no_zero | graph_zero
graph_symbols = pynini.string_file(
get_abs_path("data/electronic/symbols.tsv")).optimize()
user_name = (
pynutil.delete("username:"******"\"") +
(pynini.closure(
pynutil.add_weight(graph_digit + insert_space, 1.09)
| pynutil.add_weight(
pynini.closure(graph_symbols + pynutil.insert(" ")), 1.09)
| pynutil.add_weight(NEMO_NOT_QUOTE + insert_space, 1.1))) +
pynutil.delete("\""))
server_common = pynini.string_file(
get_abs_path("data/electronic/server_name.tsv"))
domain_common = pynini.string_file(
get_abs_path("data/electronic/domain.tsv"))
convert_defaults = (NEMO_NOT_QUOTE
| pynutil.add_weight(domain_common, -0.1)
| pynutil.add_weight(server_common, -0.1))
domain = convert_defaults + pynini.closure(
pynutil.insert(" ") + convert_defaults)
domain = pynini.compose(
domain,
pynini.closure(
pynutil.add_weight(graph_symbols, -0.1)
| pynutil.add_weight(graph_digit, -0.1) | NEMO_NOT_QUOTE),
)
domain = (pynutil.delete("domain:") + delete_space +
pynutil.delete("\"") + domain + delete_space +
pynutil.delete("\""))
protocol = pynutil.delete("protocol: \"") + pynini.closure(
NEMO_NOT_QUOTE, 1) + pynutil.delete("\"")
graph = (pynini.closure(protocol + delete_space, 0, 1) +
pynini.closure(
user_name + delete_space + pynutil.insert("at ") +
delete_space, 0, 1) + domain + delete_space)
delete_tokens = self.delete_tokens(graph)
self.fst = delete_tokens.optimize()
def get_names():
"""
Returns the graph that matched common male and female names.
"""
male_labels = load_labels(get_abs_path("data/roman/male.tsv"))
female_labels = load_labels(get_abs_path("data/roman/female.tsv"))
male_labels.extend([[x[0].upper()] for x in male_labels])
female_labels.extend([[x[0].upper()] for x in female_labels])
names = pynini.string_map(male_labels).optimize()
names |= pynini.string_map(female_labels).optimize()
return names
def get_address_graph(self, cardinal):
"""
Finite state transducer for classifying serial.
The serial is a combination of digits, letters and dashes, e.g.:
2788 San Tomas Expy, Santa Clara, CA 95051 ->
units: "address" cardinal
{ integer: "two seven eight eight San Tomas Expressway Santa Clara California nine five zero five one" }
preserve_order: true
"""
ordinal_verbalizer = OrdinalVerbalizer().graph
ordinal_tagger = OrdinalTagger(cardinal=cardinal).graph
ordinal_num = pynini.compose(
pynutil.insert("integer: \"") + ordinal_tagger +
pynutil.insert("\""), ordinal_verbalizer)
address_num = pynini.closure(NEMO_DIGIT,
1) @ cardinal.single_digits_graph
direction = (pynini.cross("E", "East")
| pynini.cross("S", "South")
| pynini.cross("W", "West")
| pynini.cross("N", "North"))
direction = pynini.closure(
pynutil.add_weight(pynini.accep(NEMO_SPACE) + direction, -1), 0, 1)
address_words = pynini.string_file(
get_abs_path("data/address/address_words.tsv"))
address_words = (pynini.accep(NEMO_SPACE) +
pynini.closure(ordinal_num, 0, 1) +
pynini.closure(NEMO_ALPHA | NEMO_SPACE, 1) +
address_words)
city = pynini.closure(NEMO_ALPHA | pynini.accep(NEMO_SPACE), 1)
city = pynini.closure(
pynini.cross(",", "") + pynini.accep(NEMO_SPACE) + city, 0, 1)
state = pynini.invert(
pynini.string_file(get_abs_path("data/address/states.tsv")))
state = pynini.closure(
pynini.cross(",", "") + pynini.accep(NEMO_SPACE) + state, 0, 1)
zip_code = pynini.compose(NEMO_DIGIT**5, cardinal.single_digits_graph)
zip_code = pynini.closure(
pynutil.add_weight(
pynini.closure(pynini.cross(",", ""), 0, 1) +
pynini.accep(NEMO_SPACE) + zip_code, -100),
0,
1,
)
address = (address_num + direction + address_words +
pynini.closure(pynini.cross(".", ""), 0, 1) + city + state +
zip_code)
return address
def __init__(self, deterministic: bool = True):
super().__init__(name="electronic", kind="verbalize", deterministic=deterministic)
graph_digit_no_zero = pynini.invert(pynini.string_file(get_abs_path("data/number/digit.tsv"))).optimize()
graph_zero = pynini.cross("0", "zero")
if not deterministic:
graph_zero |= pynini.cross("0", "o") | pynini.cross("0", "oh")
graph_digit = graph_digit_no_zero | graph_zero
graph_symbols = pynini.string_file(get_abs_path("data/electronic/symbol.tsv")).optimize()
default_chars_symbols = pynini.cdrewrite(
pynutil.insert(" ") + (graph_symbols | graph_digit) + pynutil.insert(" "), "", "", NEMO_SIGMA
)
user_name = (
pynutil.delete("username:"******"\"")
+ default_chars_symbols
+ pynutil.delete("\"")
)
domain_common = pynini.string_file(get_abs_path("data/electronic/domain.tsv"))
domain = (
default_chars_symbols
+ insert_space
+ plurals._priority_union(
domain_common, pynutil.add_weight(pynini.cross(".", "dot"), weight=0.0001), NEMO_SIGMA
)
+ pynini.closure(
insert_space + (pynini.cdrewrite(TO_UPPER, "", "", NEMO_SIGMA) @ default_chars_symbols), 0, 1
)
)
domain = (
pynutil.delete("domain:")
+ delete_space
+ pynutil.delete("\"")
+ domain
+ delete_space
+ pynutil.delete("\"")
).optimize()
protocol = pynutil.delete("protocol: \"") + pynini.closure(NEMO_NOT_QUOTE, 1) + pynutil.delete("\"")
graph = (
pynini.closure(protocol + delete_space, 0, 1)
+ pynini.closure(user_name + delete_space + pynutil.insert(" at ") + delete_space, 0, 1)
+ domain
+ delete_space
).optimize() @ pynini.cdrewrite(delete_extra_space, "", "", NEMO_SIGMA)
delete_tokens = self.delete_tokens(graph)
self.fst = delete_tokens.optimize()
def _get_whitelist_non_deterministic_graph(
file="data/whitelist_alternatives.tsv"):
whitelist = load_labels(get_abs_path(file))
whitelist_lower = [(x.lower(), y.lower()) for x, y in whitelist]
whitelist_cased = [(x, y) for x, y in whitelist]
graph = pynini.string_map(whitelist_lower + whitelist_cased)
return graph
def __init__(self, deterministic: bool = True):
super().__init__(name="punctuation",
kind="classify",
deterministic=deterministic)
s = "!#%&\'()*+,-./:;<=>?@^_`{|}~\""
punct_symbols_to_exclude = ["[", "]"]
punct_unicode = [
chr(i) for i in range(sys.maxunicode)
if category(chr(i)).startswith("P")
and chr(i) not in punct_symbols_to_exclude
]
whitelist_symbols = load_labels(
get_abs_path("data/whitelist/symbol.tsv"))
whitelist_symbols = [x[0] for x in whitelist_symbols]
self.punct_marks = [
p for p in punct_unicode + list(s) if p not in whitelist_symbols
]
punct = pynini.union(*self.punct_marks)
punct = pynini.closure(punct, 1)
emphasis = (pynini.accep("<") + (
(pynini.closure(NEMO_NOT_SPACE - pynini.union("<", ">"), 1) +
pynini.closure(pynini.accep("/"), 0, 1))
| (pynini.accep("/") +
pynini.closure(NEMO_NOT_SPACE - pynini.union("<", ">"), 1))) +
pynini.accep(">"))
punct = plurals._priority_union(emphasis, punct, NEMO_SIGMA)
self.graph = punct
self.fst = (pynutil.insert("name: \"") + self.graph +
pynutil.insert("\"")).optimize()
def get_serial_graph(self):
"""
Finite state transducer for classifying serial.
The serial is a combination of digits, letters and dashes, e.g.:
c325-b -> tokens { cardinal { integer: "c three two five b" } }
"""
alpha = NEMO_ALPHA
if self.deterministic:
num_graph = self.single_digits_graph
else:
num_graph = self.graph
letter_pronunciation = pynini.string_map(
load_labels(get_abs_path("data/letter_pronunciation.tsv")))
alpha |= letter_pronunciation
delimiter = insert_space | pynini.cross("-", " ") | pynini.cross(
"/", " ")
letter_num = pynini.closure(alpha + delimiter, 1) + num_graph
num_letter = pynini.closure(num_graph + delimiter, 1) + alpha
num_delimiter_num = pynini.closure(num_graph + delimiter,
1) + num_graph
next_alpha_or_num = pynini.closure(delimiter + (alpha | num_graph))
serial_graph = (letter_num | num_letter
| num_delimiter_num) + next_alpha_or_num
if not self.deterministic:
serial_graph += pynini.closure(
pynini.accep("s") | pynini.cross("s", "es"), 0, 1)
serial_graph.optimize()
return pynutil.add_weight(serial_graph, 10)
def __init__(self, deterministic: bool = True):
super().__init__(name="electronic",
kind="verbalize",
deterministic=deterministic)
graph_digit = pynini.invert(
pynini.string_file(
get_abs_path("data/numbers/digit.tsv"))).optimize()
graph_symbols = pynini.string_file(
get_abs_path("data/electronic/symbols.tsv")).optimize()
user_name = (
pynutil.delete("username:"******"\"") +
(pynini.closure(
pynutil.add_weight(graph_digit + insert_space, 1.09)
| pynutil.add_weight(
pynini.closure(graph_symbols + pynutil.insert(" ")), 1.09)
| pynutil.add_weight(NEMO_NOT_QUOTE + insert_space, 1.1))) +
pynutil.delete("\""))
domain_default = (pynini.closure(NEMO_NOT_QUOTE + insert_space) +
pynini.cross(".", "dot ") + NEMO_NOT_QUOTE +
pynini.closure(insert_space + NEMO_NOT_QUOTE))
server_default = (pynini.closure(
(graph_digit | NEMO_ALPHA) + insert_space, 1) +
pynini.closure(graph_symbols + insert_space) +
pynini.closure(
(graph_digit | NEMO_ALPHA) + insert_space, 1))
server_common = pynini.string_file(
get_abs_path("data/electronic/server_name.tsv")) + insert_space
domain_common = pynini.cross(".", "dot ") + pynini.string_file(
get_abs_path("data/electronic/domain.tsv"))
domain = (pynutil.delete("domain:") + delete_space +
pynutil.delete("\"") +
(pynutil.add_weight(server_common, 1.09)
| pynutil.add_weight(server_default, 1.1)) +
(pynutil.add_weight(domain_common, 1.09)
| pynutil.add_weight(domain_default, 1.1)) + delete_space +
pynutil.delete("\""))
graph = (pynini.closure(
user_name + delete_space + pynutil.insert("at ") + delete_space, 0,
1) + domain + delete_space)
delete_tokens = self.delete_tokens(graph)
self.fst = delete_tokens.optimize()
def _get_whitelist_graph(input_case, file="data/whitelist.tsv"):
whitelist = load_labels(get_abs_path(file))
if input_case == "lower_cased":
whitelist = [(x.lower(), y) for x, y in whitelist]
else:
whitelist = [(x, y) for x, y in whitelist]
graph = pynini.string_map(whitelist)
return graph
def __init__(self, deterministic: bool = True):
super().__init__(name="telephone",
kind="classify",
deterministic=deterministic)
add_separator = pynutil.insert(", ") # between components
digit = pynini.invert(
pynini.string_file(get_abs_path("data/numbers/digit.tsv"))
).optimize() | pynini.cross("0", "o")
country_code = (pynutil.insert("country_code: \"") +
pynini.closure(pynutil.delete("+"), 0, 1) +
pynini.closure(digit + insert_space, 0, 2) + digit +
pynutil.insert("\""))
optional_country_code = pynini.closure(
country_code + pynini.closure(pynutil.delete("-"), 0, 1) +
delete_space + insert_space, 0, 1)
area_part_common = pynutil.add_weight(
pynini.cross("800", "eight hundred"), -1.1)
area_part_default = pynini.closure(digit + insert_space, 2, 2) + digit
area_part = area_part_default | area_part_common
area_part = (
(area_part + pynutil.delete("-"))
| (pynutil.delete("(") + area_part +
(pynutil.delete(") ") | pynutil.delete(")-")))) + add_separator
del_separator = pynini.closure(pynini.union("-", " "), 0, 1)
number_length = ((NEMO_DIGIT + del_separator) |
(NEMO_ALPHA + del_separator))**7
number_words = pynini.closure((NEMO_DIGIT @ digit) +
(insert_space | pynini.cross("-", ', '))
| NEMO_ALPHA
| (NEMO_ALPHA + pynini.cross("-", ' ')))
number_words = pynini.compose(number_length, number_words)
number_part = area_part + number_words
number_part = pynutil.insert(
"number_part: \"") + number_part + pynutil.insert("\"")
extension = (pynutil.insert("extension: \"") +
pynini.closure(digit + insert_space, 0, 3) + digit +
pynutil.insert("\""))
optional_extension = pynini.closure(insert_space + extension, 0, 1)
graph = optional_country_code + number_part + optional_extension
# ip
digit_to_str_graph = pynini.compose(
NEMO_DIGIT**(1, 3),
digit + pynini.closure(pynutil.insert(" ") + digit)).optimize()
ip_graph = digit_to_str_graph + (pynini.cross(".", " dot ") +
digit_to_str_graph)**3
graph |= pynutil.insert(
"number_part: \"") + ip_graph.optimize() + pynutil.insert("\"")
final_graph = self.add_tokens(graph)
self.fst = final_graph.optimize()
def __init__(self, cardinal: GraphFst, deterministic: bool):
super().__init__(name="decimal",
kind="classify",
deterministic=deterministic)
cardinal_graph = cardinal.graph
cardinal_graph_hundred_component_at_least_one_none_zero_digit = (
cardinal.graph_hundred_component_at_least_one_none_zero_digit)
graph_decimal = pynini.string_file(
get_abs_path("data/numbers/digit.tsv"))
graph_decimal |= pynini.string_file(
get_abs_path("data/numbers/zero.tsv"))
graph_decimal = (
pynini.cross("zero", "0")
| graph_decimal
| (graph_decimal | pynini.cross("o", "0")) + pynini.closure(
delete_space + (graph_decimal | pynini.cross("o", "0")), 1))
self.graph = pynini.invert(graph_decimal).optimize()
if not deterministic:
self.graph = self.graph | cardinal_graph
point = pynutil.delete(".")
optional_graph_negative = pynini.closure(
pynutil.insert("negative: ") + pynini.cross("-", "\"true\" "), 0,
1)
graph_fractional = pynutil.insert(
"fractional_part: \"") + self.graph + pynutil.insert("\"")
graph_integer = pynutil.insert(
"integer_part: \"") + cardinal_graph + pynutil.insert("\"")
final_graph_wo_sign = (
pynini.closure(graph_integer + pynutil.insert(" "), 0, 1) + point +
pynutil.insert(" ") + graph_fractional)
self.final_graph_wo_negative = final_graph_wo_sign | get_quantity(
final_graph_wo_sign,
cardinal_graph_hundred_component_at_least_one_none_zero_digit)
final_graph = optional_graph_negative + self.final_graph_wo_negative
final_graph = self.add_tokens(final_graph)
self.fst = final_graph.optimize()
def __init__(self,
input_case: str,
deterministic: bool = True,
input_file: str = None):
super().__init__(name="whitelist",
kind="classify",
deterministic=deterministic)
def _get_whitelist_graph(input_case, file):
whitelist = load_labels(file)
if input_case == "lower_cased":
whitelist = [(x.lower(), y) for x, y in whitelist]
else:
whitelist = [(x, y) for x, y in whitelist]
graph = pynini.string_map(whitelist)
return graph
def _get_whitelist_non_deterministic_graph(
file="data/whitelist_alternatives.tsv"):
whitelist = load_labels(get_abs_path(file))
whitelist_lower = [(x.lower(), y.lower()) for x, y in whitelist]
whitelist_cased = [(x, y) for x, y in whitelist]
graph = pynini.string_map(whitelist_lower + whitelist_cased)
return graph
graph = _get_whitelist_graph(input_case,
get_abs_path("data/whitelist.tsv"))
if not deterministic:
graph |= (_get_whitelist_graph("lower_cased",
get_abs_path("data/whitelist.tsv"))
| _get_whitelist_non_deterministic_graph())
if input_file:
whitelist_provided = _get_whitelist_graph(input_case, input_file)
if not deterministic:
graph |= whitelist_provided
else:
graph = whitelist_provided
self.graph = (convert_space(graph)).optimize()
self.fst = (pynutil.insert("name: \"") + self.graph +
pynutil.insert("\"")).optimize()
def __init__(self,
cardinal: GraphFst,
decimal: GraphFst,
deterministic: bool = True):
super().__init__(name="money",
kind="classify",
deterministic=deterministic)
cardinal_graph = cardinal.graph
graph_decimal_final = decimal.final_graph_wo_negative
unit_singular = pynini.string_file(
get_abs_path("data/currency/currency.tsv"))
unit_plural = convert_space(unit_singular @ SINGULAR_TO_PLURAL)
unit_singular = convert_space(unit_singular)
graph_unit_singular = pynutil.insert(
"currency: \"") + unit_singular + pynutil.insert("\"")
graph_unit_plural = pynutil.insert(
"currency: \"") + unit_plural + pynutil.insert("\"")
singular_graph = (graph_unit_singular +
pynutil.insert(" integer_part: \"") +
pynini.cross("1", "one") + pynutil.insert("\""))
graph_decimal = graph_unit_plural + insert_space + graph_decimal_final
if deterministic:
graph_integer = (graph_unit_plural +
pynutil.insert(" integer_part: \"") +
((NEMO_SIGMA - "1") @ cardinal_graph) +
pynutil.insert("\""))
else:
graph_integer = (
graph_unit_plural + pynutil.insert(" integer_part: \"") +
((NEMO_SIGMA - "1")
@ (get_hundreds_graph(deterministic) | cardinal_graph)) +
pynutil.insert("\""))
graph_decimal |= singular_graph + insert_space + graph_decimal_final
graph_integer |= singular_graph
final_graph = graph_integer | graph_decimal
final_graph = self.add_tokens(final_graph)
self.fst = final_graph.optimize()
def __init__(self, deterministic: bool = True):
super().__init__(name="telephone",
kind="classify",
deterministic=deterministic)
add_separator = pynutil.insert(", ") # between components
digit = pynini.invert(
pynini.string_file(get_abs_path("data/numbers/digit.tsv"))
).optimize() | pynini.cross("0", "o")
country_code = (pynutil.insert("country_code: \"") +
pynutil.delete("+") +
pynini.closure(digit + insert_space, 0, 2) + digit +
pynutil.insert("\""))
optional_country_code = pynini.closure(
country_code + pynini.closure(pynutil.delete("-"), 0, 1) +
delete_space + insert_space, 0, 1)
number_part = ((
(pynini.closure(digit + insert_space, 2, 2) + digit +
pynutil.delete("-"))
|
(pynutil.delete("(") + pynini.closure(digit + insert_space, 2, 2) +
digit + pynutil.delete(")") +
pynini.closure(pynutil.delete("-"), 0, 1) + delete_space)) +
add_separator +
pynini.closure(digit + insert_space, 2, 2) + digit +
pynutil.delete("-") + add_separator +
pynini.closure(digit + insert_space, 3, 3) + digit)
number_part = pynutil.insert(
"number_part: \"") + number_part + pynutil.insert("\"")
extension = (pynutil.insert("extension : \"") +
pynini.closure(digit + insert_space, 0, 3) + digit +
pynutil.insert("\""))
optional_extension = pynini.closure(
insert_space + pynutil.delete("-") + extension, 0, 1)
graph = optional_country_code + number_part + optional_extension
final_graph = self.add_tokens(graph)
self.fst = final_graph.optimize()
def __init__(self,
cardinal: GraphFst,
decimal: GraphFst,
deterministic: bool = True):
super().__init__(name="money",
kind="classify",
deterministic=deterministic)
cardinal_graph = cardinal.graph_with_and
graph_decimal_final = decimal.final_graph_wo_negative_w_abbr
maj_singular_labels = load_labels(
get_abs_path("data/money/currency_major.tsv"))
maj_unit_plural = convert_space(maj_singular @ SINGULAR_TO_PLURAL)
maj_unit_singular = convert_space(maj_singular)
graph_maj_singular = pynutil.insert(
"currency_maj: \"") + maj_unit_singular + pynutil.insert("\"")
graph_maj_plural = pynutil.insert(
"currency_maj: \"") + maj_unit_plural + pynutil.insert("\"")
optional_delete_fractional_zeros = pynini.closure(
pynutil.delete(".") + pynini.closure(pynutil.delete("0"), 1), 0, 1)
graph_integer_one = pynutil.insert("integer_part: \"") + pynini.cross(
"1", "one") + pynutil.insert("\"")
# only for decimals where third decimal after comma is non-zero or with quantity
decimal_delete_last_zeros = (
pynini.closure(NEMO_DIGIT | pynutil.delete(",")) +
pynini.accep(".") + pynini.closure(NEMO_DIGIT, 2) +
(NEMO_DIGIT - "0") + pynini.closure(pynutil.delete("0")))
decimal_with_quantity = NEMO_SIGMA + NEMO_ALPHA
graph_decimal = (graph_maj_plural + insert_space +
(decimal_delete_last_zeros | decimal_with_quantity)
@ graph_decimal_final)
graph_integer = (pynutil.insert("integer_part: \"") +
((NEMO_SIGMA - "1") @ cardinal_graph) +
pynutil.insert("\""))
graph_integer_only = graph_maj_singular + insert_space + graph_integer_one
graph_integer_only |= graph_maj_plural + insert_space + graph_integer
final_graph = (graph_integer_only +
optional_delete_fractional_zeros) | graph_decimal
# remove trailing zeros of non zero number in the first 2 digits and fill up to 2 digits
# e.g. 2000 -> 20, 0200->02, 01 -> 01, 10 -> 10
# not accepted: 002, 00, 0,
two_digits_fractional_part = (
pynini.closure(NEMO_DIGIT) +
(NEMO_DIGIT - "0") + pynini.closure(pynutil.delete("0"))) @ (
(pynutil.delete("0") + (NEMO_DIGIT - "0"))
| ((NEMO_DIGIT - "0") + pynutil.insert("0"))
| ((NEMO_DIGIT - "0") + NEMO_DIGIT))
graph_min_singular = pynutil.insert(
" currency_min: \"") + min_singular + pynutil.insert("\"")
graph_min_plural = pynutil.insert(
" currency_min: \"") + min_plural + pynutil.insert("\"")
# format ** dollars ** cent
decimal_graph_with_minor = None
integer_graph_reordered = None
decimal_default_reordered = None
for curr_symbol, _ in maj_singular_labels:
preserve_order = pynutil.insert(" preserve_order: true")
integer_plus_maj = graph_integer + insert_space + pynutil.insert(
curr_symbol) @ graph_maj_plural
integer_plus_maj |= graph_integer_one + insert_space + pynutil.insert(
curr_symbol) @ graph_maj_singular
integer_plus_maj_with_comma = pynini.compose(
NEMO_DIGIT - "0" +
pynini.closure(NEMO_DIGIT | pynutil.delete(",")),
integer_plus_maj)
integer_plus_maj = pynini.compose(
pynini.closure(NEMO_DIGIT) - "0", integer_plus_maj)
integer_plus_maj |= integer_plus_maj_with_comma
graph_fractional_one = two_digits_fractional_part @ pynini.cross(
"1", "one")
graph_fractional_one = pynutil.insert(
"fractional_part: \"") + graph_fractional_one + pynutil.insert(
"\"")
graph_fractional = (two_digits_fractional_part @ (
pynini.closure(NEMO_DIGIT, 1, 2) - "1"
) @ cardinal.graph_hundred_component_at_least_one_none_zero_digit)
graph_fractional = pynutil.insert(
"fractional_part: \"") + graph_fractional + pynutil.insert(
"\"")
fractional_plus_min = graph_fractional + insert_space + pynutil.insert(
curr_symbol) @ graph_min_plural
fractional_plus_min |= (
graph_fractional_one + insert_space +
pynutil.insert(curr_symbol) @ graph_min_singular)
decimal_graph_with_minor_curr = integer_plus_maj + pynini.cross(
".", " ") + fractional_plus_min
if not deterministic:
decimal_graph_with_minor_curr |= pynutil.add_weight(
integer_plus_maj + pynini.cross(".", " ") +
pynutil.insert("fractional_part: \"") +
two_digits_fractional_part @ cardinal.
graph_hundred_component_at_least_one_none_zero_digit +
pynutil.insert("\""),
weight=0.0001,
)
default_fraction_graph = (
decimal_delete_last_zeros
| decimal_with_quantity) @ graph_decimal_final
decimal_graph_with_minor_curr |= (
pynini.closure(pynutil.delete("0"), 0, 1) +
pynutil.delete(".") + fractional_plus_min)
decimal_graph_with_minor_curr = (pynutil.delete(curr_symbol) +
decimal_graph_with_minor_curr +
preserve_order)
decimal_graph_with_minor = (
decimal_graph_with_minor_curr
if decimal_graph_with_minor is None else pynini.union(
decimal_graph_with_minor,
decimal_graph_with_minor_curr).optimize())
if not deterministic:
integer_graph_reordered_curr = (pynutil.delete(curr_symbol) +
integer_plus_maj +
preserve_order).optimize()
integer_graph_reordered = (
integer_graph_reordered_curr
if integer_graph_reordered is None else pynini.union(
integer_graph_reordered,
integer_graph_reordered_curr).optimize())
decimal_default_reordered_curr = (
pynutil.delete(curr_symbol) + default_fraction_graph +
insert_space +
pynutil.insert(curr_symbol) @ graph_maj_plural)
decimal_default_reordered = (
decimal_default_reordered_curr
if decimal_default_reordered is None else pynini.union(
decimal_default_reordered,
decimal_default_reordered_curr)).optimize()
# weight for SH
final_graph |= pynutil.add_weight(decimal_graph_with_minor, -0.0001)
if not deterministic:
final_graph |= integer_graph_reordered | decimal_default_reordered
# to handle "$2.00" cases
final_graph |= pynini.compose(
NEMO_SIGMA + pynutil.delete(".") +
pynini.closure(pynutil.delete("0"), 1),
integer_graph_reordered)
final_graph = self.add_tokens(final_graph.optimize())
self.fst = final_graph.optimize()
def __init__(self, deterministic: bool = True):
super().__init__(name="cardinal",
kind="classify",
deterministic=deterministic)
# TODO repalce to have "oh" as a default for "0"
graph = pynini.Far(
get_abs_path("data/numbers/cardinal_number_name.far")).get_fst()
self.graph_hundred_component_at_least_one_none_zero_digit = (
pynini.closure(NEMO_DIGIT, 2, 3)
| pynini.difference(NEMO_DIGIT, pynini.accep("0"))) @ graph
self.graph = (pynini.closure(NEMO_DIGIT, 1, 3) + pynini.closure(
pynini.closure(pynutil.delete(","), 0, 1) + NEMO_DIGIT +
NEMO_DIGIT + NEMO_DIGIT)) @ graph
graph_digit = pynini.string_file(
get_abs_path("data/numbers/digit.tsv"))
graph_zero = pynini.string_file(get_abs_path("data/numbers/zero.tsv"))
single_digits_graph = pynini.invert(graph_digit | graph_zero)
self.single_digits_graph = single_digits_graph + pynini.closure(
insert_space + single_digits_graph)
if not deterministic:
# for a single token allow only the same normalization
# "007" -> {"oh oh seven", "zero zero seven"} not {"oh zero seven"}
single_digits_graph_zero = pynini.invert(graph_digit | graph_zero)
single_digits_graph_oh = pynini.invert(graph_digit) | pynini.cross(
"0", "oh")
self.single_digits_graph = single_digits_graph_zero + pynini.closure(
insert_space + single_digits_graph_zero)
self.single_digits_graph |= single_digits_graph_oh + pynini.closure(
insert_space + single_digits_graph_oh)
single_digits_graph_with_commas = pynini.closure(
self.single_digits_graph + insert_space, 1,
3) + pynini.closure(
pynutil.delete(",") + single_digits_graph + insert_space +
single_digits_graph + insert_space + single_digits_graph,
1,
)
self.range_graph = pynutil.insert(
"from ") + self.graph + pynini.cross("-", " to ") + self.graph
self.range_graph |= self.graph + (pynini.cross(
"x", " by ") | pynini.cross(" x ", " by ")) + self.graph
self.range_graph |= (pynutil.insert("from ") +
get_hundreds_graph() +
pynini.cross("-", " to ") +
get_hundreds_graph())
self.range_graph = self.range_graph.optimize()
serial_graph = self.get_serial_graph()
optional_minus_graph = pynini.closure(
pynutil.insert("negative: ") + pynini.cross("-", "\"true\" "), 0,
1)
if deterministic:
long_numbers = pynini.compose(NEMO_DIGIT**(5, ...),
self.single_digits_graph).optimize()
final_graph = self.graph | serial_graph | pynutil.add_weight(
long_numbers, -0.001)
cardinal_with_leading_zeros = pynini.compose(
pynini.accep("0") + pynini.closure(NEMO_DIGIT),
self.single_digits_graph)
final_graph |= cardinal_with_leading_zeros
else:
leading_zeros = pynini.compose(
pynini.closure(pynini.accep("0"), 1), self.single_digits_graph)
cardinal_with_leading_zeros = (
leading_zeros + pynutil.insert(" ") +
pynini.compose(pynini.closure(NEMO_DIGIT), self.graph))
final_graph = (self.graph
| serial_graph
| self.range_graph
| self.single_digits_graph
| get_hundreds_graph()
| pynutil.add_weight(
single_digits_graph_with_commas, 0.001)
| cardinal_with_leading_zeros)
final_graph = optional_minus_graph + pynutil.insert(
"integer: \"") + final_graph + pynutil.insert("\"")
final_graph = self.add_tokens(final_graph)
self.fst = final_graph.optimize()
def __init__(self,
cardinal: GraphFst,
decimal: GraphFst,
fraction: GraphFst,
deterministic: bool = True):
super().__init__(name="measure",
kind="classify",
deterministic=deterministic)
cardinal_graph = cardinal.graph
if not deterministic:
cardinal_graph |= cardinal.range_graph
graph_unit = pynini.string_file(get_abs_path("data/measurements.tsv"))
graph_unit |= pynini.compose(
pynini.closure(TO_LOWER, 1) + pynini.closure(NEMO_ALPHA),
graph_unit)
graph_unit_plural = convert_space(graph_unit @ SINGULAR_TO_PLURAL)
graph_unit = convert_space(graph_unit)
optional_graph_negative = pynini.closure(
pynutil.insert("negative: ") + pynini.cross("-", "\"true\" "), 0,
1)
graph_unit2 = pynini.cross("/", "per") + delete_space + pynutil.insert(
NEMO_NON_BREAKING_SPACE) + graph_unit
optional_graph_unit2 = pynini.closure(
delete_space + pynutil.insert(NEMO_NON_BREAKING_SPACE) +
graph_unit2,
0,
1,
)
unit_plural = (
pynutil.insert("units: \"") +
(graph_unit_plural + optional_graph_unit2 | graph_unit2) +
pynutil.insert("\""))
unit_singular = (pynutil.insert("units: \"") +
(graph_unit + optional_graph_unit2 | graph_unit2) +
pynutil.insert("\""))
subgraph_decimal = (pynutil.insert("decimal { ") +
optional_graph_negative +
decimal.final_graph_wo_negative + delete_space +
pynutil.insert(" } ") + unit_plural)
subgraph_cardinal = (pynutil.insert("cardinal { ") +
optional_graph_negative +
pynutil.insert("integer: \"") +
((NEMO_SIGMA - "1") @ cardinal_graph) +
delete_space + pynutil.insert("\"") +
pynutil.insert(" } ") + unit_plural)
subgraph_cardinal |= (pynutil.insert("cardinal { ") +
optional_graph_negative +
pynutil.insert("integer: \"") +
pynini.cross("1", "one") + delete_space +
pynutil.insert("\"") + pynutil.insert(" } ") +
unit_singular)
cardinal_dash_alpha = (pynutil.insert("cardinal { integer: \"") +
cardinal_graph + pynini.accep('-') +
pynutil.insert("\" } units: \"") +
pynini.closure(NEMO_ALPHA, 1) +
pynutil.insert("\""))
alpha_dash_cardinal = (pynutil.insert("units: \"") +
pynini.closure(NEMO_ALPHA, 1) +
pynini.accep('-') + pynutil.insert("\"") +
pynutil.insert(" cardinal { integer: \"") +
cardinal_graph +
pynutil.insert("\" } preserve_order: true"))
decimal_dash_alpha = (pynutil.insert("decimal { ") +
decimal.final_graph_wo_negative +
pynini.cross('-', '') +
pynutil.insert(" } units: \"") +
pynini.closure(NEMO_ALPHA, 1) +
pynutil.insert("\""))
decimal_times = (pynutil.insert("decimal { ") +
decimal.final_graph_wo_negative +
pynutil.insert(" } units: \"") +
pynini.cross(pynini.union('x', "X"), 'x') +
pynutil.insert("\""))
alpha_dash_decimal = (pynutil.insert("units: \"") +
pynini.closure(NEMO_ALPHA, 1) +
pynini.accep('-') + pynutil.insert("\"") +
pynutil.insert(" decimal { ") +
decimal.final_graph_wo_negative +
pynutil.insert(" } preserve_order: true"))
subgraph_fraction = (pynutil.insert("fraction { ") + fraction.graph +
delete_space + pynutil.insert(" } ") +
unit_plural)
address = self.get_address_graph(cardinal)
address = (
pynutil.insert("units: \"address\" cardinal { integer: \"") +
address + pynutil.insert("\" } preserve_order: true"))
math_operations = pynini.string_file(
get_abs_path("data/math_operations.tsv"))
delimiter = pynini.accep(" ") | pynutil.insert(" ")
math = (cardinal_graph + delimiter + math_operations + delimiter +
cardinal_graph + delimiter + pynini.cross("=", "equals") +
delimiter + cardinal_graph)
math = (pynutil.insert("units: \"math\" cardinal { integer: \"") +
math + pynutil.insert("\" } preserve_order: true"))
final_graph = (subgraph_decimal
| subgraph_cardinal
| cardinal_dash_alpha
| alpha_dash_cardinal
| decimal_dash_alpha
| decimal_times
| alpha_dash_decimal
| subgraph_fraction
| address
| math)
final_graph = self.add_tokens(final_graph)
self.fst = final_graph.optimize()
def __init__(self, deterministic: bool = True, lm: bool = False):
super().__init__(name="roman",
kind="classify",
deterministic=deterministic)
roman_dict = load_labels(
get_abs_path("data/roman/roman_to_spoken.tsv"))
default_graph = pynini.string_map(roman_dict).optimize()
default_graph = pynutil.insert(
"integer: \"") + default_graph + pynutil.insert("\"")
graph_teens = pynini.string_map([x[0]
for x in roman_dict[:19]]).optimize()
# up to five digit roman numerals with a preceding name are converted to ordinal form
names = get_names()
graph = (pynutil.insert("key_the_ordinal: \"") + names +
pynutil.insert("\"") + pynini.accep(" ") +
graph_teens @ default_graph).optimize()
# single symbol roman numerals with preceding key words are converted to cardinal form
key_words = pynini.string_map(
load_labels(get_abs_path("data/roman/key_word.tsv"))).optimize()
graph |= (pynutil.insert("key_cardinal: \"") + key_words +
pynutil.insert("\"") + pynini.accep(" ") +
default_graph).optimize()
if deterministic:
# two digit roman numerals up to 49
roman_to_cardinal = pynini.compose(
pynini.closure(NEMO_ALPHA, 2),
(pynutil.insert("default_cardinal: \"default\" ") +
(pynini.string_map([x[0] for x in roman_dict[:50]
]).optimize()) @ default_graph),
)
elif not lm:
# two or more digit roman numerals
roman_to_cardinal = pynini.compose(
pynini.closure(NEMO_ALPHA, 2),
(pynutil.insert("default_cardinal: \"default\" ") +
(pynini.string_map([x[0] for x in roman_dict[:50]
]).optimize()) @ default_graph),
)
# convert three digit roman or up with suffix to ordinal
roman_to_ordinal = pynini.compose(
pynini.closure(NEMO_ALPHA, 3),
(pynutil.insert("default_ordinal: \"default\" ") +
graph_teens @ default_graph + pynutil.delete("th")),
)
graph |= roman_to_cardinal | roman_to_ordinal
# # add a higher weight when roman number consists of a single symbol
# graph = pynini.compose(pynini.closure(NEMO_CHAR, 2), graph) | pynutil.add_weight(
# pynini.compose(NEMO_CHAR, graph), 101
# )
# graph = graph.optimize() + pynini.closure(pynutil.delete("."), 0, 1)
# graph = pynutil.insert("integer: \"") + graph + pynutil.insert("\"")
graph = self.add_tokens(graph)
self.fst = graph.optimize()
def get_address_graph(self, cardinal):
"""
Finite state transducer for classifying serial.
The serial is a combination of digits, letters and dashes, e.g.:
2788 San Tomas Expy, Santa Clara, CA 95051 ->
units: "address" cardinal
{ integer: "two seven eight eight San Tomas Expressway Santa Clara California nine five zero five one" }
preserve_order: true
"""
ordinal_verbalizer = OrdinalVerbalizer().graph
ordinal_tagger = OrdinalTagger(cardinal=cardinal).graph
ordinal_num = pynini.compose(
pynutil.insert("integer: \"") + ordinal_tagger +
pynutil.insert("\""), ordinal_verbalizer)
address_num = NEMO_DIGIT**(
1,
2) @ cardinal.graph_hundred_component_at_least_one_none_zero_digit
address_num += insert_space + NEMO_DIGIT**2 @ (
pynini.closure(pynini.cross("0", "zero "), 0, 1) +
cardinal.graph_hundred_component_at_least_one_none_zero_digit)
# to handle the rest of the numbers
address_num = pynini.compose(NEMO_DIGIT**(3, 4), address_num)
address_num = plurals._priority_union(address_num, cardinal.graph,
NEMO_SIGMA)
direction = (pynini.cross("E", "East")
| pynini.cross("S", "South")
| pynini.cross("W", "West")
| pynini.cross("N", "North")) + pynini.closure(
pynutil.delete("."), 0, 1)
direction = pynini.closure(pynini.accep(NEMO_SPACE) + direction, 0, 1)
address_words = get_formats(
get_abs_path("data/address/address_word.tsv"))
address_words = (
pynini.accep(NEMO_SPACE) +
(pynini.closure(ordinal_num, 0, 1)
| NEMO_UPPER + pynini.closure(NEMO_ALPHA, 1)) + NEMO_SPACE +
pynini.closure(NEMO_UPPER + pynini.closure(NEMO_ALPHA) +
NEMO_SPACE) + address_words)
city = pynini.closure(NEMO_ALPHA | pynini.accep(NEMO_SPACE), 1)
city = pynini.closure(
pynini.accep(",") + pynini.accep(NEMO_SPACE) + city, 0, 1)
states = load_labels(get_abs_path("data/address/state.tsv"))
additional_options = []
for x, y in states:
additional_options.append((x, f"{y[0]}.{y[1:]}"))
states.extend(additional_options)
state_graph = pynini.string_map(states)
state = pynini.invert(state_graph)
state = pynini.closure(
pynini.accep(",") + pynini.accep(NEMO_SPACE) + state, 0, 1)
zip_code = pynini.compose(NEMO_DIGIT**5, cardinal.single_digits_graph)
zip_code = pynini.closure(
pynini.closure(pynini.accep(","), 0, 1) +
pynini.accep(NEMO_SPACE) + zip_code,
0,
1,
)
address = address_num + direction + address_words + pynini.closure(
city + state + zip_code, 0, 1)
address |= address_num + direction + address_words + pynini.closure(
pynini.cross(".", ""), 0, 1)
return address
NEMO_DIGIT,
NEMO_SIGMA,
TO_LOWER,
GraphFst,
delete_extra_space,
delete_space,
insert_space,
)
from nemo_text_processing.text_normalization.en.utils import get_abs_path, load_labels
try:
import pynini
from pynini.lib import pynutil
graph_teen = pynini.invert(
pynini.string_file(get_abs_path("data/numbers/teen.tsv"))).optimize()
graph_digit = pynini.invert(
pynini.string_file(get_abs_path("data/numbers/digit.tsv"))).optimize()
ties_graph = pynini.invert(
pynini.string_file(get_abs_path("data/numbers/ties.tsv"))).optimize()
PYNINI_AVAILABLE = True
except (ModuleNotFoundError, ImportError):
# Add placeholders for global variables
graph_teen = None
graph_digit = None
ties_graph = None
PYNINI_AVAILABLE = True
def __init__(self, deterministic: bool = True, lm: bool = False):
super().__init__(name="roman",
kind="classify",
deterministic=deterministic)
roman_dict = load_labels(
get_abs_path("data/roman/roman_to_spoken.tsv"))
default_graph = pynini.string_map(roman_dict).optimize()
default_graph = pynutil.insert(
"integer: \"") + default_graph + pynutil.insert("\"")
ordinal_limit = 19
graph_teens = pynini.string_map(
[x[0] for x in roman_dict[:ordinal_limit]]).optimize()
# roman numerals up to ordinal_limit with a preceding name are converted to ordinal form
names = get_names()
graph = (pynutil.insert("key_the_ordinal: \"") + names +
pynutil.insert("\"") + pynini.accep(" ") +
graph_teens @ default_graph).optimize()
# single symbol roman numerals with preceding key words (multiple formats) are converted to cardinal form
key_words = []
for k_word in load_labels(get_abs_path("data/roman/key_word.tsv")):
key_words.append(k_word)
key_words.append([k_word[0][0].upper() + k_word[0][1:]])
key_words.append([k_word[0].upper()])
key_words = pynini.string_map(key_words).optimize()
graph |= (pynutil.insert("key_cardinal: \"") + key_words +
pynutil.insert("\"") + pynini.accep(" ") +
default_graph).optimize()
if deterministic:
# two digit roman numerals up to 49
roman_to_cardinal = pynini.compose(
pynini.closure(NEMO_ALPHA, 2),
(pynutil.insert("default_cardinal: \"default\" ") +
(pynini.string_map([x[0] for x in roman_dict[:50]
]).optimize()) @ default_graph),
)
elif not lm:
# two or more digit roman numerals
roman_to_cardinal = pynini.compose(
pynini.difference(NEMO_SIGMA, "I"),
(pynutil.insert("default_cardinal: \"default\" integer: \"") +
pynini.string_map(roman_dict).optimize() +
pynutil.insert("\"")),
).optimize()
# convert three digit roman or up with suffix to ordinal
roman_to_ordinal = pynini.compose(
pynini.closure(NEMO_ALPHA, 3),
(pynutil.insert("default_ordinal: \"default\" ") +
graph_teens @ default_graph + pynutil.delete("th")),
)
graph |= roman_to_cardinal | roman_to_ordinal
graph = self.add_tokens(graph)
self.fst = graph.optimize()
def __init__(self, cardinal: GraphFst, deterministic: bool):
super().__init__(name="date",
kind="classify",
deterministic=deterministic)
month_graph = pynini.string_file(
get_abs_path("data/months/names.tsv")).optimize()
month_graph |= (TO_LOWER + pynini.closure(NEMO_CHAR)) @ month_graph
month_abbr_graph = pynini.string_file(
get_abs_path("data/months/abbr.tsv")).optimize()
month_abbr_graph = (month_abbr_graph |
(TO_LOWER + pynini.closure(NEMO_CHAR))
@ month_abbr_graph) + pynini.closure(
pynutil.delete("."), 0, 1)
month_graph |= month_abbr_graph
# to support all caps names
names_all_caps = [[
x[0].upper()
] for x in load_labels(get_abs_path("data/months/names.tsv"))]
abbr_all_caps = [
(x.upper(), y)
for x, y in load_labels(get_abs_path("data/months/abbr.tsv"))
]
month_graph |= pynini.string_map(names_all_caps) | (
pynini.string_map(abbr_all_caps) +
pynini.closure(pynutil.delete("."), 0, 1))
month_numbers_graph = pynini.string_file(
get_abs_path("data/months/numbers.tsv")).optimize()
cardinal_graph = cardinal.graph_hundred_component_at_least_one_none_zero_digit
year_graph = _get_year_graph(deterministic)
YEAR_WEIGHT = 0.001
year_graph_standalone = (pynutil.insert("year: \"") +
pynutil.add_weight(year_graph, YEAR_WEIGHT) +
pynutil.insert("\""))
month_graph = pynutil.insert(
"month: \"") + month_graph + pynutil.insert("\"")
month_numbers_graph = pynutil.insert(
"month: \"") + month_numbers_graph + pynutil.insert("\"")
day_graph = (pynutil.insert("day: \"") +
((pynini.union("1", "2", "3") + NEMO_DIGIT) | NEMO_DIGIT)
@ cardinal_graph + pynutil.insert("\""))
optional_day_graph = pynini.closure(delete_extra_space + day_graph, 0,
1)
two_digit_year = NEMO_DIGIT**(2) @ (cardinal.single_digits_graph
| cardinal_graph)
two_digit_year = pynutil.insert(
"year: \"") + two_digit_year + pynutil.insert("\"")
year_graph = pynutil.insert("year: \"") + year_graph + pynutil.insert(
"\"")
optional_graph_year = pynini.closure(
delete_extra_space + year_graph,
0,
1,
)
graph_mdy = (month_graph + optional_day_graph + delete_space +
pynini.closure(pynutil.delete(","), 0, 1) +
optional_graph_year)
delete_sep = pynutil.delete(pynini.union("-", "/", "."))
graph_mdy |= (month_numbers_graph + delete_sep + insert_space +
pynini.closure(pynutil.delete("0"), 0, 1) + day_graph +
delete_sep + insert_space +
(year_graph | two_digit_year))
graph_dmy = (day_graph + delete_extra_space + month_graph +
pynini.closure(pynutil.delete(","), 0, 1) +
optional_graph_year)
graph_ymd = ((year_graph | two_digit_year) + delete_sep +
insert_space + month_numbers_graph + delete_sep +
insert_space + pynini.closure(pynutil.delete("0"), 0, 1) +
day_graph)
final_graph = graph_mdy | graph_dmy
if deterministic:
final_graph += pynutil.insert(" preserve_order: true")
else:
final_graph += pynini.closure(
pynutil.insert(" preserve_order: true"), 0, 1)
final_graph |= graph_ymd | year_graph_standalone
if not deterministic:
ymd_to_mdy_graph = None
mdy_to_dmy_graph = None
for month in [
x[0]
for x in load_labels(get_abs_path("data/months/names.tsv"))
]:
for day in [
x[0] for x in load_labels(
get_abs_path("data/months/days.tsv"))
]:
ymd_to_mdy_curr = (
pynutil.insert("month: \"" + month + "\" day: \"" +
day + "\" ") + pynini.accep('year:') +
NEMO_SIGMA + pynutil.delete(" month: \"" + month +
"\" day: \"" + day + "\""))
# YY-MM-DD -> MM-DD-YY
ymd_to_mdy_curr = pynini.compose(final_graph,
ymd_to_mdy_curr)
ymd_to_mdy_graph = (
ymd_to_mdy_curr if ymd_to_mdy_graph is None else
pynini.union(ymd_to_mdy_curr, ymd_to_mdy_graph))
mdy_to_dmy_curr = (
pynutil.insert("day: \"" + day + "\" month: \"" +
month + "\" ") +
pynutil.delete("month: \"" + month + "\" day: \"" +
day + "\" ") + pynini.accep('year:') +
NEMO_SIGMA)
# pynini.compose(ymd_to_mdy_curr, mdy_to_dmy_curr) to handle:
# YY-MM-DD (input format) -> MM-DD-YY (intermediate ymd_to_mdy_curr representation) -> DD-MM-YY
# '2000-01-05' -> 'day: "five" month: "january" year: "two thousand"'
# pynini.compose(final_graph, mdy_to_dmy_curr) to handle:
# MM-DD-YY (input format) -> DD-MM-YY
mdy_to_dmy_curr = pynini.compose(
ymd_to_mdy_curr, mdy_to_dmy_curr) | pynini.compose(
final_graph, mdy_to_dmy_curr)
mdy_to_dmy_graph = (
mdy_to_dmy_curr if mdy_to_dmy_graph is None else
pynini.union(mdy_to_dmy_curr, mdy_to_dmy_graph))
final_graph |= ymd_to_mdy_graph | mdy_to_dmy_graph
final_graph = self.add_tokens(final_graph)
self.fst = final_graph.optimize()
NEMO_ALPHA,
NEMO_DIGIT,
NEMO_SIGMA,
SINGULAR_TO_PLURAL,
GraphFst,
convert_space,
insert_space,
)
from nemo_text_processing.text_normalization.en.utils import get_abs_path, load_labels
try:
import pynini
from pynini.lib import pynutil
min_singular = pynini.string_file(
get_abs_path("data/money/currency_minor_singular.tsv"))
min_plural = pynini.string_file(
get_abs_path("data/money/currency_minor_plural.tsv"))
maj_singular = pynini.string_file(
(get_abs_path("data/money/currency_major.tsv")))
PYNINI_AVAILABLE = True
except (ModuleNotFoundError, ImportError):
PYNINI_AVAILABLE = False
class MoneyFst(GraphFst):
"""
Finite state transducer for classifying money, suppletive aware, e.g.
$12.05 -> money { integer_part: "twelve" currency_maj: "dollars" fractional_part: "five" currency_min: "cents" preserve_order: true }
$12.0500 -> money { integer_part: "twelve" currency_maj: "dollars" fractional_part: "five" currency_min: "cents" preserve_order: true }
NEMO_PUNCT = pynini.union(
*map(pynini.escape, string.punctuation)).optimize()
NEMO_GRAPH = pynini.union(NEMO_ALNUM, NEMO_PUNCT).optimize()
NEMO_SIGMA = pynini.closure(NEMO_CHAR)
delete_space = pynutil.delete(pynini.closure(NEMO_WHITE_SPACE))
insert_space = pynutil.insert(" ")
delete_extra_space = pynini.cross(pynini.closure(NEMO_WHITE_SPACE, 1), " ")
delete_preserve_order = pynini.closure(
pynutil.delete(" preserve_order: true")
| (pynutil.delete(" field_order: \"") + NEMO_NOT_QUOTE +
pynutil.delete("\"")))
suppletive = pynini.string_file(get_abs_path("data/suppletive.tsv"))
# _v = pynini.union("a", "e", "i", "o", "u")
_c = pynini.union("b", "c", "d", "f", "g", "h", "j", "k", "l", "m", "n",
"p", "q", "r", "s", "t", "v", "w", "x", "y", "z")
_ies = NEMO_SIGMA + _c + pynini.cross("y", "ies")
_es = NEMO_SIGMA + pynini.union("s", "sh", "ch", "x",
"z") + pynutil.insert("es")
_s = NEMO_SIGMA + pynutil.insert("s")
graph_plural = plurals._priority_union(
suppletive,
plurals._priority_union(_ies,
plurals._priority_union(_es, _s, NEMO_SIGMA),
NEMO_SIGMA), NEMO_SIGMA).optimize()
SINGULAR_TO_PLURAL = graph_plural
delete_extra_space,
delete_space,
insert_space,
)
from nemo_text_processing.text_normalization.en.utils import (
augment_labels_with_punct_at_end,
get_abs_path,
load_labels,
)
try:
import pynini
from pynini.lib import pynutil
from pynini.examples import plurals
graph_teen = pynini.invert(pynini.string_file(get_abs_path("data/number/teen.tsv"))).optimize()
graph_digit = pynini.invert(pynini.string_file(get_abs_path("data/number/digit.tsv"))).optimize()
ties_graph = pynini.invert(pynini.string_file(get_abs_path("data/number/ty.tsv"))).optimize()
year_suffix = load_labels(get_abs_path("data/date/year_suffix.tsv"))
year_suffix.extend(augment_labels_with_punct_at_end(year_suffix))
year_suffix = pynini.string_map(year_suffix).optimize()
PYNINI_AVAILABLE = True
except (ModuleNotFoundError, ImportError):
# Add placeholders for global variables
graph_teen = None
graph_digit = None
ties_graph = None
PYNINI_AVAILABLE = True
def __init__(self, input_case: str, deterministic: bool = True, input_file: str = None):
super().__init__(name="whitelist", kind="classify", deterministic=deterministic)
def _get_whitelist_graph(input_case, file, keep_punct_add_end: bool = False):
whitelist = load_labels(file)
if input_case == "lower_cased":
whitelist = [[x.lower(), y] for x, y in whitelist]
else:
whitelist = [[x, y] for x, y in whitelist]
if keep_punct_add_end:
whitelist.extend(augment_labels_with_punct_at_end(whitelist))
graph = pynini.string_map(whitelist)
return graph
graph = _get_whitelist_graph(input_case, get_abs_path("data/whitelist/tts.tsv"))
graph |= _get_whitelist_graph(input_case, get_abs_path("data/whitelist/symbol.tsv"))
if deterministic:
names = get_names()
graph |= (
pynini.cross(pynini.union("st", "St", "ST"), "Saint")
+ pynini.closure(pynutil.delete("."))
+ pynini.accep(" ")
+ names
)
else:
graph |= _get_whitelist_graph(
input_case, get_abs_path("data/whitelist/alternatives.tsv"), keep_punct_add_end=True
)
for x in [".", ". "]:
graph |= (
NEMO_UPPER
+ pynini.closure(pynutil.delete(x) + NEMO_UPPER, 2)
+ pynini.closure(pynutil.delete("."), 0, 1)
)
if not deterministic:
multiple_forms_whitelist_graph = get_formats(get_abs_path("data/whitelist/alternatives_all_format.tsv"))
graph |= multiple_forms_whitelist_graph
graph_unit = pynini.string_file(get_abs_path("data/measure/unit.tsv")) | pynini.string_file(
get_abs_path("data/measure/unit_alternatives.tsv")
)
graph_unit_plural = graph_unit @ SINGULAR_TO_PLURAL
units_graph = pynini.compose(NEMO_CHAR ** (3, ...), convert_space(graph_unit | graph_unit_plural))
graph |= units_graph
# convert to states only if comma is present before the abbreviation to avoid converting all caps words,
# e.g. "IN", "OH", "OK"
# TODO or only exclude above?
states = load_labels(get_abs_path("data/address/state.tsv"))
additional_options = []
for x, y in states:
if input_case == "lower_cased":
x = x.lower()
additional_options.append((x, f"{y[0]}.{y[1:]}"))
if not deterministic:
additional_options.append((x, f"{y[0]}.{y[1:]}."))
states.extend(additional_options)
state_graph = pynini.string_map(states)
graph |= pynini.closure(NEMO_NOT_SPACE, 1) + pynini.union(", ", ",") + pynini.invert(state_graph).optimize()
if input_file:
whitelist_provided = _get_whitelist_graph(input_case, input_file)
if not deterministic:
graph |= whitelist_provided
else:
graph = whitelist_provided
self.graph = (convert_space(graph)).optimize()
self.fst = (pynutil.insert("name: \"") + self.graph + pynutil.insert("\"")).optimize()
def __init__(self, cardinal: GraphFst, deterministic: bool, lm: bool = False):
super().__init__(name="date", kind="classify", deterministic=deterministic)
# january
month_graph = pynini.string_file(get_abs_path("data/date/month_name.tsv")).optimize()
# January, JANUARY
month_graph |= pynini.compose(TO_LOWER + pynini.closure(NEMO_CHAR), month_graph) | pynini.compose(
TO_LOWER ** (2, ...), month_graph
)
# jan
month_abbr_graph = pynini.string_file(get_abs_path("data/date/month_abbr.tsv")).optimize()
# jan, Jan, JAN
month_abbr_graph = (
month_abbr_graph
| pynini.compose(TO_LOWER + pynini.closure(NEMO_LOWER, 1), month_abbr_graph).optimize()
| pynini.compose(TO_LOWER ** (2, ...), month_abbr_graph).optimize()
) + pynini.closure(pynutil.delete("."), 0, 1)
month_graph |= month_abbr_graph.optimize()
month_numbers_labels = pynini.string_file(get_abs_path("data/date/month_number.tsv")).optimize()
cardinal_graph = cardinal.graph_hundred_component_at_least_one_none_zero_digit
year_graph = _get_year_graph(cardinal_graph=cardinal_graph, deterministic=deterministic)
# three_digit_year = (NEMO_DIGIT @ cardinal_graph) + insert_space + (NEMO_DIGIT ** 2) @ cardinal_graph
# year_graph |= three_digit_year
month_graph = pynutil.insert("month: \"") + month_graph + pynutil.insert("\"")
month_numbers_graph = pynutil.insert("month: \"") + month_numbers_labels + pynutil.insert("\"")
endings = ["rd", "th", "st", "nd"]
endings += [x.upper() for x in endings]
endings = pynini.union(*endings)
day_graph = (
pynutil.insert("day: \"")
+ pynini.closure(pynutil.delete("the "), 0, 1)
+ (
((pynini.union("1", "2") + NEMO_DIGIT) | NEMO_DIGIT | (pynini.accep("3") + pynini.union("0", "1")))
+ pynini.closure(pynutil.delete(endings), 0, 1)
)
@ cardinal_graph
+ pynutil.insert("\"")
)
two_digit_year = _get_two_digit_year(
cardinal_graph=cardinal_graph, single_digits_graph=cardinal.single_digits_graph
)
two_digit_year = pynutil.insert("year: \"") + two_digit_year + pynutil.insert("\"")
# if lm:
# two_digit_year = pynini.compose(pynini.difference(NEMO_DIGIT, "0") + NEMO_DIGIT ** (3), two_digit_year)
# year_graph = pynini.compose(pynini.difference(NEMO_DIGIT, "0") + NEMO_DIGIT ** (2), year_graph)
# year_graph |= pynini.compose(pynini.difference(NEMO_DIGIT, "0") + NEMO_DIGIT ** (4, ...), year_graph)
graph_year = pynutil.insert(" year: \"") + pynutil.delete(" ") + year_graph + pynutil.insert("\"")
graph_year |= (
pynutil.insert(" year: \"")
+ pynini.accep(",")
+ pynini.closure(pynini.accep(" "), 0, 1)
+ year_graph
+ pynutil.insert("\"")
)
optional_graph_year = pynini.closure(graph_year, 0, 1)
year_graph = pynutil.insert("year: \"") + year_graph + pynutil.insert("\"")
graph_mdy = month_graph + (
(delete_extra_space + day_graph)
| (pynini.accep(" ") + day_graph)
| graph_year
| (delete_extra_space + day_graph + graph_year)
)
graph_mdy |= (
month_graph
+ pynini.cross("-", " ")
+ day_graph
+ pynini.closure(((pynini.cross("-", " ") + NEMO_SIGMA) @ graph_year), 0, 1)
)
for x in ["-", "/", "."]:
delete_sep = pynutil.delete(x)
graph_mdy |= (
month_numbers_graph
+ delete_sep
+ insert_space
+ pynini.closure(pynutil.delete("0"), 0, 1)
+ day_graph
+ delete_sep
+ insert_space
+ (year_graph | two_digit_year)
)
graph_dmy = day_graph + delete_extra_space + month_graph + optional_graph_year
day_ex_month = (NEMO_DIGIT ** 2 - pynini.project(month_numbers_graph, "input")) @ day_graph
for x in ["-", "/", "."]:
delete_sep = pynutil.delete(x)
graph_dmy |= (
day_ex_month
+ delete_sep
+ insert_space
+ month_numbers_graph
+ delete_sep
+ insert_space
+ (year_graph | two_digit_year)
)
graph_ymd = pynini.accep("")
for x in ["-", "/", "."]:
delete_sep = pynutil.delete(x)
graph_ymd |= (
(year_graph | two_digit_year)
+ delete_sep
+ insert_space
+ month_numbers_graph
+ delete_sep
+ insert_space
+ pynini.closure(pynutil.delete("0"), 0, 1)
+ day_graph
)
final_graph = graph_mdy | graph_dmy
if not deterministic or lm:
final_graph += pynini.closure(pynutil.insert(" preserve_order: true"), 0, 1)
m_sep_d = (
month_numbers_graph
+ pynutil.delete(pynini.union("-", "/"))
+ insert_space
+ pynini.closure(pynutil.delete("0"), 0, 1)
+ day_graph
)
final_graph |= m_sep_d
else:
final_graph += pynutil.insert(" preserve_order: true")
final_graph |= graph_ymd | year_graph
if not deterministic or lm:
ymd_to_mdy_graph = None
ymd_to_dmy_graph = None
mdy_to_dmy_graph = None
md_to_dm_graph = None
for month in [x[0] for x in load_labels(get_abs_path("data/date/month_name.tsv"))]:
for day in [x[0] for x in load_labels(get_abs_path("data/date/day.tsv"))]:
ymd_to_mdy_curr = (
pynutil.insert("month: \"" + month + "\" day: \"" + day + "\" ")
+ pynini.accep('year:')
+ NEMO_SIGMA
+ pynutil.delete(" month: \"" + month + "\" day: \"" + day + "\"")
)
# YY-MM-DD -> MM-DD-YY
ymd_to_mdy_curr = pynini.compose(graph_ymd, ymd_to_mdy_curr)
ymd_to_mdy_graph = (
ymd_to_mdy_curr
if ymd_to_mdy_graph is None
else pynini.union(ymd_to_mdy_curr, ymd_to_mdy_graph)
)
ymd_to_dmy_curr = (
pynutil.insert("day: \"" + day + "\" month: \"" + month + "\" ")
+ pynini.accep('year:')
+ NEMO_SIGMA
+ pynutil.delete(" month: \"" + month + "\" day: \"" + day + "\"")
)
# YY-MM-DD -> MM-DD-YY
ymd_to_dmy_curr = pynini.compose(graph_ymd, ymd_to_dmy_curr).optimize()
ymd_to_dmy_graph = (
ymd_to_dmy_curr
if ymd_to_dmy_graph is None
else pynini.union(ymd_to_dmy_curr, ymd_to_dmy_graph)
)
mdy_to_dmy_curr = (
pynutil.insert("day: \"" + day + "\" month: \"" + month + "\" ")
+ pynutil.delete("month: \"" + month + "\" day: \"" + day + "\" ")
+ pynini.accep('year:')
+ NEMO_SIGMA
).optimize()
# MM-DD-YY -> verbalize as MM-DD-YY (February fourth 1991) or DD-MM-YY (the fourth of February 1991)
mdy_to_dmy_curr = pynini.compose(graph_mdy, mdy_to_dmy_curr).optimize()
mdy_to_dmy_graph = (
mdy_to_dmy_curr
if mdy_to_dmy_graph is None
else pynini.union(mdy_to_dmy_curr, mdy_to_dmy_graph).optimize()
).optimize()
md_to_dm_curr = pynutil.insert("day: \"" + day + "\" month: \"" + month + "\"") + pynutil.delete(
"month: \"" + month + "\" day: \"" + day + "\""
)
md_to_dm_curr = pynini.compose(m_sep_d, md_to_dm_curr).optimize()
md_to_dm_graph = (
md_to_dm_curr
if md_to_dm_graph is None
else pynini.union(md_to_dm_curr, md_to_dm_graph).optimize()
).optimize()
final_graph |= mdy_to_dmy_graph | md_to_dm_graph | ymd_to_mdy_graph | ymd_to_dmy_graph
final_graph = self.add_tokens(final_graph)
self.fst = final_graph.optimize()
def __init__(self,
cardinal: GraphFst,
ordinal: GraphFst,
deterministic: bool = True,
lm: bool = False):
super().__init__(name="integer",
kind="classify",
deterministic=deterministic)
"""
Finite state transducer for classifying serial (handles only cases without delimiters,
values with delimiters are handled by default).
The serial is a combination of digits, letters and dashes, e.g.:
c325b -> tokens { cardinal { integer: "c three two five b" } }
"""
num_graph = pynini.compose(NEMO_DIGIT**(6, ...),
cardinal.single_digits_graph).optimize()
num_graph |= pynini.compose(NEMO_DIGIT**(1, 5),
cardinal.graph).optimize()
# to handle numbers starting with zero
num_graph |= pynini.compose(
pynini.accep("0") + pynini.closure(NEMO_DIGIT),
cardinal.single_digits_graph).optimize()
# TODO: "#" doesn't work from the file
symbols_graph = pynini.string_file(
get_abs_path("data/whitelist/symbol.tsv")).optimize(
) | pynini.cross("#", "hash")
num_graph |= symbols_graph
if not self.deterministic and not lm:
num_graph |= cardinal.single_digits_graph
# also allow double digits to be pronounced as integer in serial number
num_graph |= pynutil.add_weight(
NEMO_DIGIT**2 @ cardinal.
graph_hundred_component_at_least_one_none_zero_digit,
weight=0.0001)
# add space between letter and digit/symbol
symbols = [
x[0]
for x in load_labels(get_abs_path("data/whitelist/symbol.tsv"))
]
symbols = pynini.union(*symbols)
digit_symbol = NEMO_DIGIT | symbols
graph_with_space = pynini.compose(
pynini.cdrewrite(pynutil.insert(" "), NEMO_ALPHA | symbols,
digit_symbol, NEMO_SIGMA),
pynini.cdrewrite(pynutil.insert(" "), digit_symbol,
NEMO_ALPHA | symbols, NEMO_SIGMA),
)
# serial graph with delimiter
delimiter = pynini.accep("-") | pynini.accep("/") | pynini.accep(" ")
alphas = pynini.closure(NEMO_ALPHA, 1)
letter_num = alphas + delimiter + num_graph
num_letter = pynini.closure(num_graph + delimiter, 1) + alphas
next_alpha_or_num = pynini.closure(delimiter + (alphas | num_graph))
next_alpha_or_num |= pynini.closure(
delimiter + num_graph +
plurals._priority_union(pynini.accep(" "), pynutil.insert(" "),
NEMO_SIGMA).optimize() + alphas)
serial_graph = letter_num + next_alpha_or_num
serial_graph |= num_letter + next_alpha_or_num
# numbers only with 2+ delimiters
serial_graph |= (num_graph + delimiter + num_graph + delimiter +
num_graph + pynini.closure(delimiter + num_graph))
# 2+ symbols
serial_graph |= pynini.compose(NEMO_SIGMA + symbols + NEMO_SIGMA,
num_graph + delimiter + num_graph)
# exclude ordinal numbers from serial options
serial_graph = pynini.compose(
pynini.difference(NEMO_SIGMA,
pynini.project(ordinal.graph, "input")),
serial_graph).optimize()
serial_graph = pynutil.add_weight(serial_graph, 0.0001)
serial_graph |= (pynini.closure(NEMO_NOT_SPACE, 1) +
(pynini.cross("^2", " squared")
| pynini.cross("^3", " cubed")).optimize())
# at least one serial graph with alpha numeric value and optional additional serial/num/alpha values
serial_graph = (
pynini.closure((serial_graph | num_graph | alphas) + delimiter) +
serial_graph + pynini.closure(delimiter +
(serial_graph | num_graph | alphas)))
serial_graph |= pynini.compose(graph_with_space,
serial_graph.optimize()).optimize()
serial_graph = pynini.compose(pynini.closure(NEMO_NOT_SPACE, 2),
serial_graph).optimize()
self.graph = serial_graph.optimize()
graph = pynutil.insert("name: \"") + convert_space(
self.graph).optimize() + pynutil.insert("\"")
self.fst = graph.optimize()