diff --git a/Jenkinsfile b/Jenkinsfile index fd7a2c988..b1696c262 100644 --- a/Jenkinsfile +++ b/Jenkinsfile @@ -10,7 +10,6 @@ pipeline { disableConcurrentBuilds(abortPrevious: true) } environment { - AR_TN_CACHE='/home/jenkins/TestData/text_norm/ci/grammars/04-24-24-0' DE_TN_CACHE='/home/jenkins/TestData/text_norm/ci/grammars/10-23-24-0' EN_TN_CACHE='/home/jenkins/TestData/text_norm/ci/grammars/09-25-25-0' @@ -27,7 +26,7 @@ pipeline { HY_TN_CACHE='/home/jenkins/TestData/text_norm/ci/grammars/03-12-24-0' MR_TN_CACHE='/home/jenkins/TestData/text_norm/ci/grammars/03-12-24-1' JA_TN_CACHE='/home/jenkins/TestData/text_norm/ci/grammars/10-17-24-1' - HI_TN_CACHE='/home/jenkins/TestData/text_norm/ci/grammars/04-22-25-0' + HI_TN_CACHE='/home/jenkins/TestData/text_norm/ci/grammars/10-31-25-0' DEFAULT_TN_CACHE='/home/jenkins/TestData/text_norm/ci/grammars/06-08-23-0' } stages { diff --git a/nemo_text_processing/text_normalization/hi/data/measure/quarterly_units_list.tsv b/nemo_text_processing/text_normalization/hi/data/measure/quarterly_units_list.tsv new file mode 100644 index 000000000..6fcfb8b3a --- /dev/null +++ b/nemo_text_processing/text_normalization/hi/data/measure/quarterly_units_list.tsv @@ -0,0 +1,5 @@ +हफ़्ते +सप्ताह +सदियां +सदियों + diff --git a/nemo_text_processing/text_normalization/hi/data/measure/quarterly_units.tsv b/nemo_text_processing/text_normalization/hi/data/measure/quarterly_units_map.tsv similarity index 77% rename from nemo_text_processing/text_normalization/hi/data/measure/quarterly_units.tsv rename to nemo_text_processing/text_normalization/hi/data/measure/quarterly_units_map.tsv index eaddf930a..dc20bcb21 100644 --- a/nemo_text_processing/text_normalization/hi/data/measure/quarterly_units.tsv +++ b/nemo_text_processing/text_normalization/hi/data/measure/quarterly_units_map.tsv @@ -4,9 +4,8 @@ h घंटे min मिनट doz दर्जन yr साल -yr वर्ष hp हॉर्सपॉवर d दिन month महीना months महीने -हफ़्ते हफ़्ते \ No newline at end of file + diff --git a/nemo_text_processing/text_normalization/hi/data/measure/unit.tsv b/nemo_text_processing/text_normalization/hi/data/measure/unit.tsv index 189512687..4065bc86b 100644 --- a/nemo_text_processing/text_normalization/hi/data/measure/unit.tsv +++ b/nemo_text_processing/text_normalization/hi/data/measure/unit.tsv @@ -134,7 +134,6 @@ KHz किलोहर्ट्ज़ N न्यूटन dB डेसीबल yr साल -yr वर्ष hp हॉर्सपॉवर d दिन month महीना diff --git a/nemo_text_processing/text_normalization/hi/data/ordinal/__init__.py b/nemo_text_processing/text_normalization/hi/data/ordinal/__init__.py new file mode 100644 index 000000000..341a77c5b --- /dev/null +++ b/nemo_text_processing/text_normalization/hi/data/ordinal/__init__.py @@ -0,0 +1,13 @@ +# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. diff --git a/nemo_text_processing/text_normalization/hi/data/ordinal/exceptions.tsv b/nemo_text_processing/text_normalization/hi/data/ordinal/exceptions.tsv new file mode 100644 index 000000000..bfe5738d0 --- /dev/null +++ b/nemo_text_processing/text_normalization/hi/data/ordinal/exceptions.tsv @@ -0,0 +1,12 @@ +१ला पहला +१ली पहली +२रा दूसरा +२री दूसरी +३रा तीसरा +३री तीसरी +४था चौथा +४थी चौथी +५वां पाँचवां +५वीं पाँचवीं +६ठा छठा +६ठी छठी diff --git a/nemo_text_processing/text_normalization/hi/data/ordinal/suffixes.tsv b/nemo_text_processing/text_normalization/hi/data/ordinal/suffixes.tsv new file mode 100644 index 000000000..922e9d6b8 --- /dev/null +++ b/nemo_text_processing/text_normalization/hi/data/ordinal/suffixes.tsv @@ -0,0 +1,3 @@ +वां +वीं +वें diff --git a/nemo_text_processing/text_normalization/hi/data/ordinal/suffixes_map.tsv b/nemo_text_processing/text_normalization/hi/data/ordinal/suffixes_map.tsv new file mode 100644 index 000000000..77139cff5 --- /dev/null +++ b/nemo_text_processing/text_normalization/hi/data/ordinal/suffixes_map.tsv @@ -0,0 +1,2 @@ +वे वें + diff --git a/nemo_text_processing/text_normalization/hi/data/telephone/__init__.py b/nemo_text_processing/text_normalization/hi/data/telephone/__init__.py new file mode 100644 index 000000000..341a77c5b --- /dev/null +++ b/nemo_text_processing/text_normalization/hi/data/telephone/__init__.py @@ -0,0 +1,13 @@ +# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. diff --git a/nemo_text_processing/text_normalization/hi/data/telephone/credit_context.tsv b/nemo_text_processing/text_normalization/hi/data/telephone/credit_context.tsv new file mode 100644 index 000000000..46b485af6 --- /dev/null +++ b/nemo_text_processing/text_normalization/hi/data/telephone/credit_context.tsv @@ -0,0 +1,3 @@ +नंबर +कार्ड +क्रेडिट \ No newline at end of file diff --git a/nemo_text_processing/text_normalization/hi/data/telephone/landline_context.tsv b/nemo_text_processing/text_normalization/hi/data/telephone/landline_context.tsv new file mode 100644 index 000000000..17a123bee --- /dev/null +++ b/nemo_text_processing/text_normalization/hi/data/telephone/landline_context.tsv @@ -0,0 +1,5 @@ +नंबर +मोबाइल +फोन +लैंडलाइन +कॉल \ No newline at end of file diff --git a/nemo_text_processing/text_normalization/hi/data/telephone/mobile_context.tsv b/nemo_text_processing/text_normalization/hi/data/telephone/mobile_context.tsv new file mode 100644 index 000000000..f2fa6e52f --- /dev/null +++ b/nemo_text_processing/text_normalization/hi/data/telephone/mobile_context.tsv @@ -0,0 +1,4 @@ +नंबर +मोबाइल +फोन +कॉल \ No newline at end of file diff --git a/nemo_text_processing/text_normalization/hi/data/telephone/number.tsv b/nemo_text_processing/text_normalization/hi/data/telephone/number.tsv new file mode 100644 index 000000000..e8c04b723 --- /dev/null +++ b/nemo_text_processing/text_normalization/hi/data/telephone/number.tsv @@ -0,0 +1,10 @@ +0 शून्य +1 एक +2 दो +3 तीन +4 चार +5 पाँच +6 छह +7 सात +8 आठ +9 नौ \ No newline at end of file diff --git a/nemo_text_processing/text_normalization/hi/data/telephone/pincode_context.tsv b/nemo_text_processing/text_normalization/hi/data/telephone/pincode_context.tsv new file mode 100644 index 000000000..322c7248e --- /dev/null +++ b/nemo_text_processing/text_normalization/hi/data/telephone/pincode_context.tsv @@ -0,0 +1,4 @@ +नंबर +पिन +कोड +पिनकोड \ No newline at end of file diff --git a/nemo_text_processing/text_normalization/hi/data/whitelist/paune_mappings.tsv b/nemo_text_processing/text_normalization/hi/data/whitelist/paune_mappings.tsv new file mode 100644 index 000000000..3477871e4 --- /dev/null +++ b/nemo_text_processing/text_normalization/hi/data/whitelist/paune_mappings.tsv @@ -0,0 +1,100 @@ +० एक +१ दो +२ तीन +३ चार +४ पाँच +५ छह +६ सात +७ आठ +८ नौ +९ दस +१० ग्यारह +११ बारह +१२ तेरह +१३ चौदह +१४ पंद्रह +१५ सोलह +१६ सत्रह +१७ अठारह +१८ उन्नीस +१९ बीस +२० इक्कीस +२१ बाईस +२२ तेईस +२३ चौबीस +२४ पच्चीस +२५ छब्बीस +२६ सत्ताईस +२७ अट्ठाईस +२८ उनतीस +२९ तीस +३० इकतीस +३१ बत्तीस +३२ तैंतीस +३३ चौंतीस +३४ पैंतीस +३५ छत्तीस +३६ सैंतीस +३७ अड़तीस +३८ उनतालीस +३९ चालीस +४० इकतालीस +४१ बयालीस +४२ तैंतालीस +४३ चौवालीस +४४ पैंतालीस +४५ छियालीस +४६ सैंतालीस +४७ अड़तालीस +४८ उनचास +४९ पचास +५० इक्यावन +५१ बावन +५२ तिरेपन +५३ चौवन +५४ पचपन +५५ छप्पन +५६ सत्तावन +५७ अट्ठावन +५८ उनसठ +५९ साठ +६० इकसठ +६१ बासठ +६२ तिरेसठ +६३ चौंसठ +६४ पैंसठ +६५ छियासठ +६६ सड़सठ +६७ अड़सठ +६८ उनहत्तर +६९ सत्तर +७० इकहत्तर +७१ बहत्तर +७२ तिहत्तर +७३ चौहत्तर +७४ पचहत्तर +७५ छिहत्तर +७६ सतहत्तर +७७ अठहत्तर +७८ उनासी +७९ अस्सी +८० इक्यासी +८१ बयासी +८२ तिरासी +८३ चौरासी +८४ पचासी +८५ छियासी +८६ सत्तासी +८७ अट्ठासी +८८ नवासी +८९ नब्बे +९० इक्यानबे +९१ बानबे +९२ तिरानबे +९३ चौरानबे +९४ पंचानबे +९५ छियानबे +९६ सत्तानबे +९७ अट्ठानबे +९८ निन्यानबे +९९ एक सौ diff --git a/nemo_text_processing/text_normalization/hi/graph_utils.py b/nemo_text_processing/text_normalization/hi/graph_utils.py index 6a5d3c699..5bbc736fd 100644 --- a/nemo_text_processing/text_normalization/hi/graph_utils.py +++ b/nemo_text_processing/text_normalization/hi/graph_utils.py @@ -30,6 +30,13 @@ NEMO_HI_DIGIT = pynini.union("०", "१", "२", "३", "४", "५", "६", "७", "८", "९").optimize() NEMO_HI_NON_ZERO = pynini.union("१", "२", "३", "४", "५", "६", "७", "८", "९").optimize() NEMO_HI_ZERO = "०" + +HI_DEDH = "डेढ़" # 1.5 +HI_DHAI = "ढाई" # 2.5 +HI_SAVVA = "सवा" # quarter more (1.25) +HI_SADHE = "साढ़े" # half more (X.5) +HI_PAUNE = "पौने" # quarter less (0.75) + NEMO_LOWER = pynini.union(*string.ascii_lowercase).optimize() NEMO_UPPER = pynini.union(*string.ascii_uppercase).optimize() NEMO_ALPHA = pynini.union(NEMO_LOWER, NEMO_UPPER).optimize() diff --git a/nemo_text_processing/text_normalization/hi/taggers/cardinal.py b/nemo_text_processing/text_normalization/hi/taggers/cardinal.py index c50384acf..f361416f4 100644 --- a/nemo_text_processing/text_normalization/hi/taggers/cardinal.py +++ b/nemo_text_processing/text_normalization/hi/taggers/cardinal.py @@ -15,18 +15,18 @@ import pynini from pynini.lib import pynutil -from nemo_text_processing.text_normalization.hi.graph_utils import GraphFst +from nemo_text_processing.text_normalization.hi.graph_utils import GraphFst, insert_space from nemo_text_processing.text_normalization.hi.utils import get_abs_path class CardinalFst(GraphFst): """ - Finite state transducer for classifying cardinals, e.g. - -२३ -> cardinal { negative: "true" integer: "तेइस" } } - s - Args: - deterministic: if True will provide a single transduction option, - for False multiple transduction are generated (used for audio-based normalization) + Finite state transducer for classifying cardinals, e.g. + -२३ -> cardinal { negative: "true" integer: "तेइस" } + + Args: + deterministic: if True will provide a single transduction option, + for False multiple transduction are generated (used for audio-based normalization) """ def __init__(self, deterministic: bool = True, lm: bool = False): @@ -37,6 +37,10 @@ def __init__(self, deterministic: bool = True, lm: bool = False): teens_ties = pynini.string_file(get_abs_path("data/numbers/teens_and_ties.tsv")) teens_and_ties = pynutil.add_weight(teens_ties, -0.1) + self.digit = digit + self.zero = zero + self.teens_and_ties = teens_and_ties + def create_graph_suffix(digit_graph, suffix, zeros_counts): zero = pynutil.add_weight(pynutil.delete("०"), -0.1) if zeros_counts == 0: @@ -294,6 +298,12 @@ def create_larger_number_graph(digit_graph, suffix, zeros_counts, sub_graph): graph_ten_shankhs |= create_larger_number_graph(teens_and_ties, suffix_shankhs, 0, graph_ten_padmas) graph_ten_shankhs.optimize() + # Only match exactly 2 digits to avoid interfering with telephone numbers, decimals, etc. + # e.g., "०५" -> "शून्य पाँच" + single_digit = digit | zero + graph_leading_zero = zero + insert_space + single_digit + graph_leading_zero = pynutil.add_weight(graph_leading_zero, 0.5) + final_graph = ( digit | zero @@ -315,6 +325,7 @@ def create_larger_number_graph(digit_graph, suffix, zeros_counts, sub_graph): | graph_ten_padmas | graph_shankhs | graph_ten_shankhs + | graph_leading_zero ) optional_minus_graph = pynini.closure(pynutil.insert("negative: ") + pynini.cross("-", "\"true\" "), 0, 1) diff --git a/nemo_text_processing/text_normalization/hi/taggers/date.py b/nemo_text_processing/text_normalization/hi/taggers/date.py index 37b192165..b25abcac6 100644 --- a/nemo_text_processing/text_normalization/hi/taggers/date.py +++ b/nemo_text_processing/text_normalization/hi/taggers/date.py @@ -65,11 +65,11 @@ def __init__(self, cardinal: GraphFst): (NEMO_HI_DIGIT + NEMO_HI_NON_ZERO + NEMO_HI_DIGIT + NEMO_HI_DIGIT), cardinal.graph_hundreds_as_thousand ) - cardinal_graph = ( - digit | teens_and_ties | cardinal.graph_hundreds | graph_year_thousands | graph_year_hundreds_as_thousands + cardinal_graph = pynini.union( + digit, teens_and_ties, cardinal.graph_hundreds, graph_year_thousands, graph_year_hundreds_as_thousands ) - graph_year = graph_year_thousands | graph_year_hundreds_as_thousands + graph_year = pynini.union(graph_year_thousands, graph_year_hundreds_as_thousands) delete_dash = pynutil.delete("-") delete_slash = pynutil.delete("/") @@ -102,13 +102,10 @@ def __init__(self, cardinal: GraphFst): # Updated logic to use prefix_union year_prefix = pynutil.insert("era: \"") + prefix_union + insert_space + graph_year + pynutil.insert("\"") - graph_dd_mm_yyyy = ( - days_graph + (delete_dash | delete_slash) + months_graph + (delete_dash | delete_slash) + years_graph - ) + delete_separator = pynini.union(delete_dash, delete_slash) + graph_dd_mm_yyyy = days_graph + delete_separator + months_graph + delete_separator + years_graph - graph_mm_dd_yyyy = ( - months_graph + (delete_dash | delete_slash) + days_graph + (delete_dash | delete_slash) + years_graph - ) + graph_mm_dd_yyyy = months_graph + delete_separator + days_graph + delete_separator + years_graph graph_mm_dd_yyyy += pynutil.insert(" preserve_order: true ") diff --git a/nemo_text_processing/text_normalization/hi/taggers/decimal.py b/nemo_text_processing/text_normalization/hi/taggers/decimal.py index 955e8c0d3..cb21d85b1 100644 --- a/nemo_text_processing/text_normalization/hi/taggers/decimal.py +++ b/nemo_text_processing/text_normalization/hi/taggers/decimal.py @@ -58,9 +58,7 @@ class DecimalFst(GraphFst): def __init__(self, cardinal: GraphFst, deterministic: bool = True): super().__init__(name="decimal", kind="classify", deterministic=deterministic) - graph_digit = pynini.string_file(get_abs_path("data/numbers/digit.tsv")) - graph_digit |= pynini.string_file(get_abs_path("data/numbers/zero.tsv")) - + graph_digit = cardinal.digit | cardinal.zero cardinal_graph = cardinal.final_graph self.graph = graph_digit + pynini.closure(insert_space + graph_digit).optimize() diff --git a/nemo_text_processing/text_normalization/hi/taggers/fraction.py b/nemo_text_processing/text_normalization/hi/taggers/fraction.py index 8971cd3dd..b5528deba 100644 --- a/nemo_text_processing/text_normalization/hi/taggers/fraction.py +++ b/nemo_text_processing/text_normalization/hi/taggers/fraction.py @@ -15,7 +15,20 @@ import pynini from pynini.lib import pynutil -from nemo_text_processing.text_normalization.hi.graph_utils import GraphFst +from nemo_text_processing.text_normalization.hi.graph_utils import ( + HI_DEDH, + HI_DHAI, + HI_PAUNE, + HI_SADHE, + HI_SAVVA, + NEMO_SPACE, + GraphFst, +) +from nemo_text_processing.text_normalization.hi.utils import get_abs_path + +HI_ONE_HALF = "१/२" # 1/2 +HI_ONE_QUARTER = "१/४" # 1/4 +HI_THREE_QUARTERS = "३/४" # 3/4 class FractionFst(GraphFst): @@ -39,21 +52,76 @@ def __init__(self, cardinal, deterministic: bool = True): cardinal_graph = cardinal.final_graph self.optional_graph_negative = pynini.closure( - pynutil.insert("negative: ") + pynini.cross("-", "\"true\" "), 0, 1 + pynutil.insert("negative: ") + pynini.cross("-", "\"true\"") + pynutil.insert(NEMO_SPACE), 0, 1 ) self.integer = pynutil.insert("integer_part: \"") + cardinal_graph + pynutil.insert("\"") self.numerator = ( - pynutil.insert("numerator: \"") + cardinal_graph + pynini.cross(pynini.union("/", " / "), "\" ") + pynutil.insert("numerator: \"") + + cardinal_graph + + pynini.cross(pynini.union("/", NEMO_SPACE + "/" + NEMO_SPACE), "\"") + + pynutil.insert(NEMO_SPACE) ) self.denominator = pynutil.insert("denominator: \"") + cardinal_graph + pynutil.insert("\"") - self.graph = ( + dedh_dhai_graph = pynini.string_map( + [("१" + NEMO_SPACE + HI_ONE_HALF, HI_DEDH), ("२" + NEMO_SPACE + HI_ONE_HALF, HI_DHAI)] + ) + + savva_numbers = cardinal_graph + pynini.cross(NEMO_SPACE + HI_ONE_QUARTER, "") + savva_graph = pynutil.insert(HI_SAVVA) + pynutil.insert(NEMO_SPACE) + savva_numbers + + sadhe_numbers = cardinal_graph + pynini.cross(NEMO_SPACE + HI_ONE_HALF, "") + sadhe_graph = pynutil.insert(HI_SADHE) + pynutil.insert(NEMO_SPACE) + sadhe_numbers + + paune = pynini.string_file(get_abs_path("data/whitelist/paune_mappings.tsv")) + paune_numbers = paune + pynini.cross(NEMO_SPACE + HI_THREE_QUARTERS, "") + paune_graph = pynutil.insert(HI_PAUNE) + pynutil.insert(NEMO_SPACE) + paune_numbers + + graph_dedh_dhai = ( + pynutil.insert("morphosyntactic_features: \"") + + dedh_dhai_graph + + pynutil.insert("\"") + + pynutil.insert(NEMO_SPACE) + ) + + graph_savva = ( + pynutil.insert("morphosyntactic_features: \"") + + savva_graph + + pynutil.insert("\"") + + pynutil.insert(NEMO_SPACE) + ) + + graph_sadhe = ( + pynutil.insert("morphosyntactic_features: \"") + + sadhe_graph + + pynutil.insert("\"") + + pynutil.insert(NEMO_SPACE) + ) + + graph_paune = ( + pynutil.insert("morphosyntactic_features: \"") + + paune_graph + + pynutil.insert("\"") + + pynutil.insert(NEMO_SPACE) + ) + + final_graph = ( self.optional_graph_negative - + pynini.closure(self.integer + pynini.accep(" "), 0, 1) + + pynini.closure(self.integer + pynini.accep(NEMO_SPACE), 0, 1) + self.numerator + self.denominator ) + weighted_graph = ( + final_graph + | pynutil.add_weight(graph_dedh_dhai, -0.2) + | pynutil.add_weight(graph_savva, -0.1) + | pynutil.add_weight(graph_sadhe, -0.1) + | pynutil.add_weight(graph_paune, -0.2) + ) + + self.graph = weighted_graph + graph = self.graph - final_graph = self.add_tokens(graph) - self.fst = final_graph.optimize() + graph = self.add_tokens(graph) + self.fst = graph.optimize() diff --git a/nemo_text_processing/text_normalization/hi/taggers/measure.py b/nemo_text_processing/text_normalization/hi/taggers/measure.py index 9f1ffbd39..b7d74731e 100644 --- a/nemo_text_processing/text_normalization/hi/taggers/measure.py +++ b/nemo_text_processing/text_normalization/hi/taggers/measure.py @@ -15,9 +15,24 @@ import pynini from pynini.lib import pynutil -from nemo_text_processing.text_normalization.hi.graph_utils import GraphFst, delete_space, insert_space +from nemo_text_processing.text_normalization.hi.graph_utils import ( + HI_DEDH, + HI_DHAI, + HI_PAUNE, + HI_SADHE, + HI_SAVVA, + NEMO_SPACE, + GraphFst, + delete_space, + insert_space, +) from nemo_text_processing.text_normalization.hi.utils import get_abs_path +HI_POINT_FIVE = ".५" # .5 +HI_ONE_POINT_FIVE = "१.५" # 1.5 +HI_TWO_POINT_FIVE = "२.५" # 2.5 +HI_DECIMAL_25 = ".२५" # .25 +HI_DECIMAL_75 = ".७५" # .75 digit = pynini.string_file(get_abs_path("data/numbers/digit.tsv")) teens_ties = pynini.string_file(get_abs_path("data/numbers/teens_and_ties.tsv")) @@ -41,8 +56,9 @@ def __init__(self, cardinal: GraphFst, decimal: GraphFst): super().__init__(name="measure", kind="classify") cardinal_graph = ( - digit - | teens_and_ties + cardinal.zero + | cardinal.digit + | cardinal.teens_and_ties | cardinal.graph_hundreds | cardinal.graph_thousands | cardinal.graph_ten_thousands @@ -53,7 +69,11 @@ def __init__(self, cardinal: GraphFst, decimal: GraphFst): decimal_integers = pynutil.insert("integer_part: \"") + cardinal_graph + pynutil.insert("\"") decimal_graph = decimal_integers + point + insert_space + decimal.graph_fractional unit_graph = pynini.string_file(get_abs_path("data/measure/unit.tsv")) - quarterly_units_graph = pynini.string_file(get_abs_path("data/measure/quarterly_units.tsv")) + + # Load quarterly units from separate files: map (FST) and list (FSA) + quarterly_units_map = pynini.string_file(get_abs_path("data/measure/quarterly_units_map.tsv")) + quarterly_units_list = pynini.string_file(get_abs_path("data/measure/quarterly_units_list.tsv")) + quarterly_units_graph = pynini.union(quarterly_units_map, quarterly_units_list) optional_graph_negative = pynini.closure( pynutil.insert("negative: ") + pynini.cross("-", "\"true\"") + insert_space, @@ -64,16 +84,28 @@ def __init__(self, cardinal: GraphFst, decimal: GraphFst): # Define the quarterly measurements quarter = pynini.string_map( [ - (".५", "साढ़े"), - ("१.५", "डेढ़"), - ("२.५", "ढाई"), + (HI_POINT_FIVE, HI_SADHE), + (HI_ONE_POINT_FIVE, HI_DEDH), + (HI_TWO_POINT_FIVE, HI_DHAI), ] ) quarter_graph = pynutil.insert("integer_part: \"") + quarter + pynutil.insert("\"") # Define the unit handling - unit = pynutil.insert(" units: \"") + unit_graph + pynutil.insert("\" ") - units = pynutil.insert(" units: \"") + quarterly_units_graph + pynutil.insert("\" ") + unit = ( + pynutil.insert(NEMO_SPACE) + + pynutil.insert("units: \"") + + unit_graph + + pynutil.insert("\"") + + pynutil.insert(NEMO_SPACE) + ) + units = ( + pynutil.insert(NEMO_SPACE) + + pynutil.insert("units: \"") + + quarterly_units_graph + + pynutil.insert("\"") + + pynutil.insert(NEMO_SPACE) + ) # Handling symbols like x, X, * symbol_graph = pynini.string_map( @@ -93,10 +125,71 @@ def __init__(self, cardinal: GraphFst, decimal: GraphFst): + unit ) - graph_quarter = ( + dedh_dhai = pynini.string_map([(HI_ONE_POINT_FIVE, HI_DEDH), (HI_TWO_POINT_FIVE, HI_DHAI)]) + dedh_dhai_graph = pynutil.insert("integer: \"") + dedh_dhai + pynutil.insert("\"") + + savva_numbers = cardinal_graph + pynini.cross(HI_DECIMAL_25, "") + savva_graph = ( + pynutil.insert("integer: \"") + + pynutil.insert(HI_SAVVA) + + pynutil.insert(NEMO_SPACE) + + savva_numbers + + pynutil.insert("\"") + ) + + sadhe_numbers = cardinal_graph + pynini.cross(HI_POINT_FIVE, "") + sadhe_graph = ( + pynutil.insert("integer: \"") + + pynutil.insert(HI_SADHE) + + pynutil.insert(NEMO_SPACE) + + sadhe_numbers + + pynutil.insert("\"") + ) + + paune = pynini.string_file(get_abs_path("data/whitelist/paune_mappings.tsv")) + paune_numbers = paune + pynini.cross(HI_DECIMAL_75, "") + paune_graph = ( + pynutil.insert("integer: \"") + + pynutil.insert(HI_PAUNE) + + pynutil.insert(NEMO_SPACE) + + paune_numbers + + pynutil.insert("\"") + ) + + graph_dedh_dhai = ( pynutil.insert("cardinal { ") + optional_graph_negative - + quarter_graph + + dedh_dhai_graph + + pynutil.insert(NEMO_SPACE) + + pynutil.insert("}") + + delete_space + + units + ) + + graph_savva = ( + pynutil.insert("cardinal { ") + + optional_graph_negative + + savva_graph + + pynutil.insert(NEMO_SPACE) + + pynutil.insert("}") + + delete_space + + units + ) + + graph_sadhe = ( + pynutil.insert("cardinal { ") + + optional_graph_negative + + sadhe_graph + + pynutil.insert(NEMO_SPACE) + + pynutil.insert("}") + + delete_space + + units + ) + + graph_paune = ( + pynutil.insert("cardinal { ") + + optional_graph_negative + + paune_graph + pynutil.insert(" }") + delete_space + units @@ -108,7 +201,8 @@ def __init__(self, cardinal: GraphFst, decimal: GraphFst): + pynutil.insert("integer: \"") + cardinal_graph + pynutil.insert("\"") - + pynutil.insert(" }") + + pynutil.insert(NEMO_SPACE) + + pynutil.insert("}") + delete_space + unit ) @@ -121,9 +215,11 @@ def __init__(self, cardinal: GraphFst, decimal: GraphFst): + cardinal_graph + pynutil.insert("\"") + pynutil.insert(" }") - + pynutil.insert(" units: \"") + + pynutil.insert(NEMO_SPACE) + + pynutil.insert("units: \"") + symbol_graph - + pynutil.insert("\" ") + + pynutil.insert("\"") + + pynutil.insert(NEMO_SPACE) + pynutil.insert("} }") + insert_space + pynutil.insert("tokens { cardinal { ") @@ -134,10 +230,13 @@ def __init__(self, cardinal: GraphFst, decimal: GraphFst): ) graph = ( - pynutil.add_weight(graph_decimal, 0.01) - | pynutil.add_weight(graph_quarter, 0.005) - | pynutil.add_weight(graph_cardinal, 0.01) - | pynutil.add_weight(graph_exceptions, 0.01) + pynutil.add_weight(graph_decimal, 0.1) + | pynutil.add_weight(graph_cardinal, 0.1) + | pynutil.add_weight(graph_exceptions, 0.1) + | pynutil.add_weight(graph_dedh_dhai, -0.2) + | pynutil.add_weight(graph_savva, -0.1) + | pynutil.add_weight(graph_sadhe, -0.1) + | pynutil.add_weight(graph_paune, -0.5) ) self.graph = graph.optimize() diff --git a/nemo_text_processing/text_normalization/hi/taggers/ordinal.py b/nemo_text_processing/text_normalization/hi/taggers/ordinal.py new file mode 100644 index 000000000..5f1cefed4 --- /dev/null +++ b/nemo_text_processing/text_normalization/hi/taggers/ordinal.py @@ -0,0 +1,49 @@ +# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import pynini +from pynini.lib import pynutil + +from nemo_text_processing.text_normalization.hi.graph_utils import GraphFst +from nemo_text_processing.text_normalization.hi.taggers.cardinal import CardinalFst +from nemo_text_processing.text_normalization.hi.utils import get_abs_path + + +class OrdinalFst(GraphFst): + """ + Finite state transducer for classifying Hindi ordinals, e.g. + १०वां -> ordinal { integer: "दसवां" } + २१वीं -> ordinal { integer: "इक्कीसवीं" } + + Args: + deterministic: if True will provide a single transduction option, + for False multiple transduction are generated (used for audio-based normalization) + """ + + def __init__(self, cardinal: CardinalFst, deterministic: bool = True): + super().__init__(name="ordinal", kind="classify", deterministic=deterministic) + + suffixes_list = pynini.string_file(get_abs_path("data/ordinal/suffixes.tsv")) + suffixes_map = pynini.string_file(get_abs_path("data/ordinal/suffixes_map.tsv")) + suffixes_fst = pynini.union(suffixes_list, suffixes_map) + exceptions = pynini.string_file(get_abs_path("data/ordinal/exceptions.tsv")) + + graph = cardinal.final_graph + suffixes_fst + exceptions = pynutil.add_weight(exceptions, -0.1) + graph = pynini.union(exceptions, graph) + + final_graph = pynutil.insert("integer: \"") + graph + pynutil.insert("\"") + final_graph = self.add_tokens(final_graph) + + self.fst = final_graph.optimize() diff --git a/nemo_text_processing/text_normalization/hi/taggers/punctuation.py b/nemo_text_processing/text_normalization/hi/taggers/punctuation.py index 8309ba030..14c9a1a55 100644 --- a/nemo_text_processing/text_normalization/hi/taggers/punctuation.py +++ b/nemo_text_processing/text_normalization/hi/taggers/punctuation.py @@ -36,9 +36,9 @@ def __init__(self, deterministic: bool = True): 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.union( + (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(">") ) diff --git a/nemo_text_processing/text_normalization/hi/taggers/telephone.py b/nemo_text_processing/text_normalization/hi/taggers/telephone.py new file mode 100644 index 000000000..d20870c0d --- /dev/null +++ b/nemo_text_processing/text_normalization/hi/taggers/telephone.py @@ -0,0 +1,228 @@ +# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + + +import pynini +from pynini.lib import pynutil + +from nemo_text_processing.text_normalization.hi.graph_utils import ( + NEMO_CHAR, + NEMO_DIGIT, + NEMO_HI_DIGIT, + NEMO_SPACE, + NEMO_WHITE_SPACE, + GraphFst, + delete_space, + insert_space, +) +from nemo_text_processing.text_normalization.hi.utils import get_abs_path + +HI_ZERO_DIGIT = pynini.union("0", "०") +HI_MOBILE_START_DIGITS = pynini.union("६", "७", "८", "९", "6", "7", "8", "9").optimize() +HI_LANDLINE_START_DIGITS = pynini.union("२", "३", "४", "६", "2", "3", "4", "6").optimize() + +delete_zero = pynutil.delete(HI_ZERO_DIGIT) +delete_zero_optional = pynini.closure(delete_zero, 0, 1) +insert_shunya = pynutil.insert('शून्य') + insert_space + +# Load the number mappings from the TSV file +digit_to_word = pynini.string_file(get_abs_path("data/telephone/number.tsv")) +digits = pynini.string_file(get_abs_path("data/numbers/digit.tsv")) +zero = pynini.string_file(get_abs_path("data/numbers/zero.tsv")) +mobile_context = pynini.string_file(get_abs_path("data/telephone/mobile_context.tsv")) +landline_context = pynini.string_file(get_abs_path("data/telephone/landline_context.tsv")) +credit_context = pynini.string_file(get_abs_path("data/telephone/credit_context.tsv")) +pincode_context = pynini.string_file(get_abs_path("data/telephone/pincode_context.tsv")) + +# Reusable optimized graph for any digit token +num_token = pynini.union(digit_to_word, digits, zero).optimize() + + +def generate_mobile(context_keywords: pynini.Fst) -> pynini.Fst: + context_before, context_after = get_context(context_keywords) + + # Filter cardinals to only include allowed digits + mobile_start_digit = pynini.union(HI_MOBILE_START_DIGITS @ digits, HI_MOBILE_START_DIGITS @ digit_to_word) + + country_code_digits = pynini.closure(num_token + insert_space, 1, 3) + country_code = ( + pynutil.insert("country_code: \"") + + context_before + + pynini.cross("+", "प्लस") + + insert_space + + country_code_digits + + pynutil.insert("\" ") + + pynini.closure(delete_space, 0, 1) + ) + + extension_optional = pynini.closure( + pynutil.insert("extension: \"") + + pynini.closure(num_token + insert_space, 1, 3) + + context_after + + pynutil.insert("\" ") + + delete_space, + 0, + 1, + ) + + number_part = mobile_start_digit + insert_space + pynini.closure(num_token + insert_space, 9) + + number_without_country = ( + pynutil.insert("number_part: \"") + + context_before + + delete_zero_optional + + insert_shunya + + number_part + + context_after + + pynutil.insert("\" ") + + delete_space + ) + + number_with_country = ( + country_code + + pynutil.insert("number_part: \"") + + number_part + + context_after + + pynutil.insert("\" ") + + delete_space + ) + + return (pynini.union(number_with_country, number_without_country) + extension_optional).optimize() + + +def get_landline(std_length: int, context_keywords: pynini.Fst) -> pynini.Fst: + context_before, context_after = get_context(context_keywords) + + # Filter cardinals to only include allowed digits + landline_start_digit = pynini.union(HI_LANDLINE_START_DIGITS @ digits, HI_LANDLINE_START_DIGITS @ digit_to_word) + + std_code_graph = ( + delete_zero_optional + insert_shunya + pynini.closure(num_token + insert_space, std_length, std_length) + ) + + landline_digit_count = 9 - std_length + landline_graph = ( + landline_start_digit + + insert_space + + pynini.closure(num_token + insert_space, landline_digit_count, landline_digit_count) + ) + + separator_optional = pynini.closure(pynini.union(pynini.cross("-", ""), pynini.cross(".", "")), 0, 1) + + std_code_in_brackets = ( + delete_zero_optional + + delete_space + + pynutil.delete("(") + + pynini.closure(delete_space, 0, 1) + + std_code_graph + + pynini.closure(delete_space, 0, 1) + + pynutil.delete(")") + ) + + std_part = pynini.union(std_code_graph, std_code_in_brackets) + + return ( + pynutil.insert("number_part: \"") + + context_before + + std_part + + separator_optional + + delete_space + + landline_graph + + context_after + + pynutil.insert("\" ") + ).optimize() + + +def generate_landline(context_keywords: pynini.Fst) -> pynini.Fst: + graph = ( + get_landline(2, context_keywords) + | get_landline(3, context_keywords) + | get_landline(4, context_keywords) + | get_landline(5, context_keywords) + | get_landline(6, context_keywords) + | get_landline(7, context_keywords) + ) + + return graph.optimize() + + +def get_context(keywords: pynini.Fst): + + all_digits = pynini.union(NEMO_HI_DIGIT, NEMO_DIGIT) + + non_digit_char = pynini.difference(NEMO_CHAR, pynini.union(all_digits, NEMO_WHITE_SPACE)) + word = pynini.closure(non_digit_char, 1) + pynini.accep(NEMO_SPACE) + + window = pynini.closure(word, 0, 5) + + before = pynini.closure(keywords + pynini.accep(NEMO_SPACE) + window, 0, 1) + + after = pynini.closure(pynutil.delete(NEMO_SPACE) + window + keywords, 0, 1) + + return before.optimize(), after.optimize() + + +def generate_credit(context_keywords: pynini.Fst) -> pynini.Fst: + context_before, context_after = get_context(context_keywords) + return ( + pynutil.insert("number_part: \"") + + context_before + + pynini.closure(num_token + insert_space, 4) + + context_after + + pynutil.insert("\" ") + + delete_space + ).optimize() + + +def generate_pincode(context_keywords: pynini.Fst) -> pynini.Fst: + context_before, context_after = get_context(context_keywords) + return ( + pynutil.insert("number_part: \"") + + context_before + + pynini.closure(num_token + insert_space, 6) + + context_after + + pynutil.insert("\" ") + + delete_space + ).optimize() + + +class TelephoneFst(GraphFst): + """ + Finite state transducer for tagging telephone numbers, e.g. + ९१५७११४००७ -> telephone { number_part: "शून्य नौ एक पाँच सात एक एक चार शून्य शून्य सात" } + +९१ ९२१०५१५६०६ -> telephone { country_code: "प्लस नौ एक", number_part: "नौ दो एक शून्य पाँच एक पाँच छह शून्य छह" } + १३७४-३०९९८८ -> telephone { number_part: "शून्य एक तीन सात चार तीन शून्य नौ नौ आठ आठ" } + + Args: + deterministic: if True will provide a single transduction option, + for False multiple transduction are generated (used for audio-based normalization + """ + + def __init__(self): + super().__init__(name="telephone", kind="classify") + + mobile_number = generate_mobile(mobile_context) + landline = generate_landline(landline_context) + credit_card = generate_credit(credit_context) + pincode = generate_pincode(pincode_context) + + graph = ( + pynutil.add_weight(mobile_number, 0.7) + | pynutil.add_weight(landline, 0.8) + | pynutil.add_weight(credit_card, 0.9) + | pynutil.add_weight(pincode, 1) + ) + + self.final = graph.optimize() + self.fst = self.add_tokens(self.final) diff --git a/nemo_text_processing/text_normalization/hi/taggers/time.py b/nemo_text_processing/text_normalization/hi/taggers/time.py index 6c87c9aad..09defaab2 100644 --- a/nemo_text_processing/text_normalization/hi/taggers/time.py +++ b/nemo_text_processing/text_normalization/hi/taggers/time.py @@ -15,9 +15,24 @@ import pynini from pynini.lib import pynutil -from nemo_text_processing.text_normalization.hi.graph_utils import GraphFst, insert_space +from nemo_text_processing.text_normalization.hi.graph_utils import ( + HI_DEDH, + HI_DHAI, + HI_PAUNE, + HI_SADHE, + HI_SAVVA, + NEMO_SPACE, + GraphFst, + insert_space, +) from nemo_text_processing.text_normalization.hi.utils import get_abs_path +# Time patterns specific to time tagger +HI_DOUBLE_ZERO = "००" +HI_TIME_FIFTEEN = ":१५" # :15 +HI_TIME_THIRTY = ":३०" # :30 +HI_TIME_FORTYFIVE = ":४५" # :45 + hours_graph = pynini.string_file(get_abs_path("data/time/hours.tsv")) minutes_graph = pynini.string_file(get_abs_path("data/time/minutes.tsv")) seconds_graph = pynini.string_file(get_abs_path("data/time/seconds.tsv")) @@ -36,10 +51,11 @@ class TimeFst(GraphFst): for False multiple transduction are generated (used for audio-based normalization) """ - def __init__(self): + def __init__(self, cardinal: GraphFst): super().__init__(name="time", kind="classify") delete_colon = pynutil.delete(":") + cardinal_graph = cardinal.digit | cardinal.teens_and_ties self.hours = pynutil.insert("hours: \"") + hours_graph + pynutil.insert("\" ") self.minutes = pynutil.insert("minutes: \"") + minutes_graph + pynutil.insert("\" ") @@ -54,9 +70,57 @@ def __init__(self): graph_hm = self.hours + delete_colon + insert_space + self.minutes # hour - graph_h = self.hours + delete_colon + pynutil.delete("००") + graph_h = self.hours + delete_colon + pynutil.delete(HI_DOUBLE_ZERO) + + dedh_dhai_graph = pynini.string_map([("१" + HI_TIME_THIRTY, HI_DEDH), ("२" + HI_TIME_THIRTY, HI_DHAI)]) + + savva_numbers = cardinal_graph + pynini.cross(HI_TIME_FIFTEEN, "") + savva_graph = pynutil.insert(HI_SAVVA) + pynutil.insert(NEMO_SPACE) + savva_numbers + + sadhe_numbers = cardinal_graph + pynini.cross(HI_TIME_THIRTY, "") + sadhe_graph = pynutil.insert(HI_SADHE) + pynutil.insert(NEMO_SPACE) + sadhe_numbers + + paune = pynini.string_file(get_abs_path("data/whitelist/paune_mappings.tsv")) + paune_numbers = paune + pynini.cross(HI_TIME_FORTYFIVE, "") + paune_graph = pynutil.insert(HI_PAUNE) + pynutil.insert(NEMO_SPACE) + paune_numbers + + graph_dedh_dhai = ( + pynutil.insert("morphosyntactic_features: \"") + + dedh_dhai_graph + + pynutil.insert("\"") + + pynutil.insert(NEMO_SPACE) + ) - final_graph = graph_hms | graph_hm | graph_h + graph_savva = ( + pynutil.insert("morphosyntactic_features: \"") + + savva_graph + + pynutil.insert("\"") + + pynutil.insert(NEMO_SPACE) + ) + + graph_sadhe = ( + pynutil.insert("morphosyntactic_features: \"") + + sadhe_graph + + pynutil.insert("\"") + + pynutil.insert(NEMO_SPACE) + ) + + graph_paune = ( + pynutil.insert("morphosyntactic_features: \"") + + paune_graph + + pynutil.insert("\"") + + pynutil.insert(NEMO_SPACE) + ) + + final_graph = ( + graph_hms + | pynutil.add_weight(graph_hm, 0.3) + | pynutil.add_weight(graph_h, 0.3) + | pynutil.add_weight(graph_dedh_dhai, 0.1) + | pynutil.add_weight(graph_savva, 0.2) + | pynutil.add_weight(graph_sadhe, 0.2) + | pynutil.add_weight(graph_paune, 0.1) + ) final_graph = self.add_tokens(final_graph) self.fst = final_graph.optimize() diff --git a/nemo_text_processing/text_normalization/hi/taggers/tokenize_and_classify.py b/nemo_text_processing/text_normalization/hi/taggers/tokenize_and_classify.py index b1bbd2a10..e3e6fc5d8 100644 --- a/nemo_text_processing/text_normalization/hi/taggers/tokenize_and_classify.py +++ b/nemo_text_processing/text_normalization/hi/taggers/tokenize_and_classify.py @@ -14,12 +14,12 @@ import logging import os -import time import pynini from pynini.lib import pynutil from nemo_text_processing.text_normalization.hi.graph_utils import ( + NEMO_SPACE, NEMO_WHITE_SPACE, GraphFst, delete_extra_space, @@ -32,7 +32,9 @@ from nemo_text_processing.text_normalization.hi.taggers.fraction import FractionFst from nemo_text_processing.text_normalization.hi.taggers.measure import MeasureFst from nemo_text_processing.text_normalization.hi.taggers.money import MoneyFst +from nemo_text_processing.text_normalization.hi.taggers.ordinal import OrdinalFst from nemo_text_processing.text_normalization.hi.taggers.punctuation import PunctuationFst +from nemo_text_processing.text_normalization.hi.taggers.telephone import TelephoneFst from nemo_text_processing.text_normalization.hi.taggers.time import TimeFst from nemo_text_processing.text_normalization.hi.taggers.whitelist import WhiteListFst from nemo_text_processing.text_normalization.hi.taggers.word import WordFst @@ -77,51 +79,39 @@ def __init__( else: logging.info(f"Creating ClassifyFst grammars.") - start_time = time.time() cardinal = CardinalFst(deterministic=deterministic) cardinal_graph = cardinal.fst - logging.debug(f"cardinal: {time.time() - start_time: .2f}s -- {cardinal_graph.num_states()} nodes") - start_time = time.time() decimal = DecimalFst(cardinal=cardinal, deterministic=deterministic) decimal_graph = decimal.fst - logging.debug(f"decimal: {time.time() - start_time: .2f}s -- {decimal_graph.num_states()} nodes") - start_time = time.time() fraction = FractionFst(cardinal=cardinal, deterministic=deterministic) fraction_graph = fraction.fst - logging.debug(f"fraction: {time.time() - start_time: .2f}s -- {fraction_graph.num_states()} nodes") - start_time = time.time() date = DateFst(cardinal=cardinal) date_graph = date.fst - logging.debug(f"date: {time.time() - start_time: .2f}s -- {date_graph.num_states()} nodes") - start_time = time.time() - timefst = TimeFst() + timefst = TimeFst(cardinal=cardinal) time_graph = timefst.fst - logging.debug(f"time: {time.time() - start_time: .2f}s -- {time_graph.num_states()} nodes") - start_time = time.time() measure = MeasureFst(cardinal=cardinal, decimal=decimal) measure_graph = measure.fst - logging.debug(f"measure: {time.time() - start_time: .2f}s -- {measure_graph.num_states()} nodes") - start_time = time.time() money = MoneyFst(cardinal=cardinal) money_graph = money.fst - logging.debug(f"money: {time.time() - start_time: .2f}s -- {money_graph.num_states()} nodes") - start_time = time.time() + ordinal = OrdinalFst(cardinal=cardinal, deterministic=deterministic) + ordinal_graph = ordinal.fst + whitelist_graph = WhiteListFst( input_case=input_case, deterministic=deterministic, input_file=whitelist ).fst - logging.debug(f"whitelist: {time.time() - start_time: .2f}s -- {whitelist_graph.num_states()} nodes") - start_time = time.time() punctuation = PunctuationFst(deterministic=deterministic) punct_graph = punctuation.fst - logging.debug(f"punct: {time.time() - start_time: .2f}s -- {punct_graph.num_states()} nodes") + + telephone = TelephoneFst() + telephone_graph = telephone.fst classify = ( pynutil.add_weight(whitelist_graph, 1.01) @@ -132,35 +122,39 @@ def __init__( | pynutil.add_weight(time_graph, 1.1) | pynutil.add_weight(measure_graph, 1.1) | pynutil.add_weight(money_graph, 1.1) + | pynutil.add_weight(telephone_graph, 1.1) + | pynutil.add_weight(ordinal_graph, 1.1) ) - start_time = time.time() word_graph = WordFst(punctuation=punctuation, deterministic=deterministic).fst - logging.debug(f"word: {time.time() - start_time: .2f}s -- {word_graph.num_states()} nodes") punct = pynutil.insert("tokens { ") + pynutil.add_weight(punct_graph, weight=2.1) + pynutil.insert(" }") punct = pynini.closure( - pynini.compose(pynini.closure(NEMO_WHITE_SPACE, 1), delete_extra_space) - | (pynutil.insert(" ") + punct), + pynini.union( + pynini.compose(pynini.closure(NEMO_WHITE_SPACE, 1), delete_extra_space), + (pynutil.insert(NEMO_SPACE) + punct), + ), 1, ) - classify |= pynutil.add_weight(word_graph, 100) + classify = pynini.union(classify, pynutil.add_weight(word_graph, 100)) token = pynutil.insert("tokens { ") + classify + pynutil.insert(" }") token_plus_punct = ( - pynini.closure(punct + pynutil.insert(" ")) + token + pynini.closure(pynutil.insert(" ") + punct) + pynini.closure(punct + pynutil.insert(NEMO_SPACE)) + + token + + pynini.closure(pynutil.insert(NEMO_SPACE) + punct) ) graph = token_plus_punct + pynini.closure( - ( - pynini.compose(pynini.closure(NEMO_WHITE_SPACE, 1), delete_extra_space) - | (pynutil.insert(" ") + punct + pynutil.insert(" ")) + pynini.union( + pynini.compose(pynini.closure(NEMO_WHITE_SPACE, 1), delete_extra_space), + (pynutil.insert(NEMO_SPACE) + punct + pynutil.insert(NEMO_SPACE)), ) + token_plus_punct ) graph = delete_space + graph + delete_space - graph |= punct + graph = pynini.union(graph, punct) self.fst = graph.optimize() diff --git a/nemo_text_processing/text_normalization/hi/taggers/word.py b/nemo_text_processing/text_normalization/hi/taggers/word.py index bc354232b..00feb1827 100644 --- a/nemo_text_processing/text_normalization/hi/taggers/word.py +++ b/nemo_text_processing/text_normalization/hi/taggers/word.py @@ -40,10 +40,9 @@ def __init__(self, punctuation: PunctuationFst, deterministic: bool = True): # Define Hindi characters and symbols using pynini.union HINDI_CHAR = pynini.union( - *[chr(i) for i in range(ord("ऀ"), ord("ः") + 1)], # Hindi vowels and consonants - *[chr(i) for i in range(ord("अ"), ord("ह") + 1)], # More Hindi characters - *[chr(i) for i in range(ord("ा"), ord("्") + 1)], # Hindi diacritics - *[chr(i) for i in range(ord("०"), ord("९") + 1)], # Hindi digits + *[chr(i) for i in range(0x0900, 0x0903 + 1)], # Hindi vowels and consonants + *[chr(i) for i in range(0x0905, 0x0939 + 1)], # More Hindi characters + *[chr(i) for i in range(0x093E, 0x094D + 1)], # Hindi diacritics ).optimize() # Include punctuation in the graph diff --git a/nemo_text_processing/text_normalization/hi/verbalizers/fraction.py b/nemo_text_processing/text_normalization/hi/verbalizers/fraction.py index 7e3b33b7c..a07c41eae 100644 --- a/nemo_text_processing/text_normalization/hi/verbalizers/fraction.py +++ b/nemo_text_processing/text_normalization/hi/verbalizers/fraction.py @@ -40,6 +40,9 @@ def __init__(self, cardinal: GraphFst, deterministic: bool = True): denominator = pynutil.delete("denominator: \"") + pynini.closure(NEMO_NOT_QUOTE) + pynutil.delete("\"") insert_bata = pynutil.insert(" बटा ") insert_aur = pynutil.insert(" और ") + graph_quarter = ( + pynutil.delete("morphosyntactic_features: \"") + pynini.closure(NEMO_NOT_QUOTE, 1) + pynutil.delete("\"") + ) fraction_default = numerator + insert_bata + denominator @@ -47,7 +50,7 @@ def __init__(self, cardinal: GraphFst, deterministic: bool = True): optional_sign + pynini.closure(pynini.closure(integer, 0, 1) + insert_space + insert_aur) + fraction_default - ) + ) | graph_quarter graph = self.graph diff --git a/nemo_text_processing/text_normalization/hi/verbalizers/ordinal.py b/nemo_text_processing/text_normalization/hi/verbalizers/ordinal.py new file mode 100644 index 000000000..ab88603f6 --- /dev/null +++ b/nemo_text_processing/text_normalization/hi/verbalizers/ordinal.py @@ -0,0 +1,38 @@ +# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import pynini +from pynini.lib import pynutil + +from nemo_text_processing.text_normalization.hi.graph_utils import NEMO_NOT_QUOTE, GraphFst, delete_space + + +class OrdinalFst(GraphFst): + """ + Finite state transducer for verbalizing Hindi ordinals, e.g. + ordinal { integer: "दसवां" } -> दसवां + ordinal { integer: "इक्कीसवीं" } -> इक्कीसवीं + + Args: + deterministic: if True will provide a single transduction option, + for False multiple options (used for audio-based normalization) + """ + + def __init__(self, deterministic: bool = True): + super().__init__(name="ordinal", kind="verbalize", deterministic=deterministic) + + integer_value = delete_space + pynutil.delete("\"") + pynini.closure(NEMO_NOT_QUOTE, 1) + pynutil.delete("\"") + graph = pynutil.delete("integer:") + integer_value + delete_tokens = self.delete_tokens(graph) + self.fst = delete_tokens.optimize() diff --git a/nemo_text_processing/text_normalization/hi/verbalizers/telephone.py b/nemo_text_processing/text_normalization/hi/verbalizers/telephone.py new file mode 100644 index 000000000..55ebeab01 --- /dev/null +++ b/nemo_text_processing/text_normalization/hi/verbalizers/telephone.py @@ -0,0 +1,72 @@ +# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + + +import pynini +from pynini.lib import pynutil + +from nemo_text_processing.text_normalization.hi.graph_utils import ( + MIN_NEG_WEIGHT, + NEMO_NOT_QUOTE, + NEMO_SPACE, + GraphFst, + delete_space, + insert_space, +) + + +class TelephoneFst(GraphFst): + """ + Finite state transducer for verbalizing telephone numbers, e.g. + telephone { country_code: "प्लस नौ एक", number_part: "नौ दो एक शून्य पाँच एक पाँच छह शून्य छह" } -> प्लस नौ एक नौ दो एक शून्य पाँच एक पाँच छह शून्य छह + telephone { number_part: "शून्य एक तीन सात चार तीन शून्य नौ नौ आठ आठ" } -> शून्य एक तीन सात चार तीन शून्य नौ नौ आठ आठ + + Args: + deterministic: if True will provide a single transduction option, + for False multiple transduction are generated (used for audio-based normalization) + """ + + def __init__(self, deterministic: bool = True): + super().__init__(name="telephone", kind="verbalize", deterministic=deterministic) + + optional_country_code = pynini.closure( + pynutil.delete("country_code: \"") + + pynini.closure(NEMO_NOT_QUOTE, 1) + + pynutil.delete("\"") + + delete_space + + insert_space, + 0, + 1, + ) + + number_part = ( + pynutil.delete("number_part: \"") + + pynini.closure(NEMO_NOT_QUOTE, 1) + + pynini.closure(pynutil.add_weight(pynutil.delete(NEMO_SPACE), MIN_NEG_WEIGHT), 0, 1) + + pynutil.delete("\"") + ) + + optional_extension = pynini.closure( + delete_space + + insert_space + + pynutil.delete("extension: \"") + + pynini.closure(NEMO_NOT_QUOTE, 1) + + pynutil.delete("\""), + 0, + 1, + ) + + graph = optional_country_code + number_part + optional_extension + delete_tokens = self.delete_tokens(graph) + self.fst = delete_tokens.optimize() diff --git a/nemo_text_processing/text_normalization/hi/verbalizers/time.py b/nemo_text_processing/text_normalization/hi/verbalizers/time.py index da10df4a0..df232e3cd 100644 --- a/nemo_text_processing/text_normalization/hi/verbalizers/time.py +++ b/nemo_text_processing/text_normalization/hi/verbalizers/time.py @@ -30,7 +30,7 @@ class TimeFst(GraphFst): for False multiple transduction are generated (used for audio-based normalization) """ - def __init__(self): + def __init__(self, cardinal: GraphFst): super().__init__(name="time", kind="verbalize") hour = pynutil.delete("hours: \"") + pynini.closure(NEMO_NOT_QUOTE, 1) + pynutil.delete("\"") + insert_space @@ -63,13 +63,17 @@ def __init__(self): + insert_second ) + graph_quarter = ( + pynutil.delete("morphosyntactic_features: \"") + pynini.closure(NEMO_NOT_QUOTE, 1) + pynutil.delete("\"") + ) + # hour minute graph_hm = hour + delete_space + insert_bajkar + insert_space + minute + delete_space + insert_minute # hour graph_h = hour + delete_space + insert_baje - self.graph = graph_hms | graph_hm | graph_h + self.graph = graph_hms | graph_hm | graph_h | graph_quarter final_graph = self.graph diff --git a/nemo_text_processing/text_normalization/hi/verbalizers/verbalize.py b/nemo_text_processing/text_normalization/hi/verbalizers/verbalize.py index e91f0d9f6..12ae316b1 100644 --- a/nemo_text_processing/text_normalization/hi/verbalizers/verbalize.py +++ b/nemo_text_processing/text_normalization/hi/verbalizers/verbalize.py @@ -19,6 +19,8 @@ from nemo_text_processing.text_normalization.hi.verbalizers.fraction import FractionFst from nemo_text_processing.text_normalization.hi.verbalizers.measure import MeasureFst from nemo_text_processing.text_normalization.hi.verbalizers.money import MoneyFst +from nemo_text_processing.text_normalization.hi.verbalizers.ordinal import OrdinalFst +from nemo_text_processing.text_normalization.hi.verbalizers.telephone import TelephoneFst from nemo_text_processing.text_normalization.hi.verbalizers.time import TimeFst from nemo_text_processing.text_normalization.hi.verbalizers.whitelist import WhiteListFst @@ -49,7 +51,7 @@ def __init__(self, deterministic: bool = True): date = DateFst() date_graph = date.fst - time = TimeFst() + time = TimeFst(cardinal=cardinal) time_graph = time.fst measure = MeasureFst(cardinal=cardinal, decimal=decimal) @@ -58,6 +60,11 @@ def __init__(self, deterministic: bool = True): money = MoneyFst() money_graph = money.fst + telephone = TelephoneFst() + telephone_graph = telephone.fst + ordinal = OrdinalFst(deterministic=deterministic) + ordinal_graph = ordinal.fst + whitelist_graph = WhiteListFst(deterministic=deterministic).fst graph = ( @@ -68,7 +75,9 @@ def __init__(self, deterministic: bool = True): | time_graph | measure_graph | money_graph + | ordinal_graph | whitelist_graph + | telephone_graph ) self.fst = graph diff --git a/tests/nemo_text_processing/hi/data_text_normalization/test_cases_cardinal.txt b/tests/nemo_text_processing/hi/data_text_normalization/test_cases_cardinal.txt index 6ba21de69..2a52b2a20 100644 --- a/tests/nemo_text_processing/hi/data_text_normalization/test_cases_cardinal.txt +++ b/tests/nemo_text_processing/hi/data_text_normalization/test_cases_cardinal.txt @@ -143,3 +143,5 @@ ११०२२३४५५६७~ग्यारह अरब दो करोड़ तेईस लाख पैंतालीस हज़ार पाँच सौ सड़सठ ५१०२२३४५५६७~इक्यावन अरब दो करोड़ तेईस लाख पैंतालीस हज़ार पाँच सौ सड़सठ २ पॉइंट्स १२ गोल~दो पॉइंट्स बारह गोल +०५~शून्य पाँच +०१~शून्य एक \ No newline at end of file diff --git a/tests/nemo_text_processing/hi/data_text_normalization/test_cases_ordinal.txt b/tests/nemo_text_processing/hi/data_text_normalization/test_cases_ordinal.txt new file mode 100644 index 000000000..9bdcab2a4 --- /dev/null +++ b/tests/nemo_text_processing/hi/data_text_normalization/test_cases_ordinal.txt @@ -0,0 +1,62 @@ +१ला~पहला +१ली~पहली +२रा~दूसरा +२री~दूसरी +३रा~तीसरा +३री~तीसरी +४था~चौथा +४थी~चौथी +५वां~पाँचवां +५वीं~पाँचवीं +६ठा~छठा +६ठी~छठी +७वां~सातवां +७वीं~सातवीं +८वां~आठवां +८वीं~आठवीं +९वां~नौवां +९वीं~नौवीं +११वां~ग्यारहवां +१२वीं~बारहवीं +१४वां~चौदहवां +१६वीं~सोलहवीं +१७वां~सत्रहवां +१८वीं~अठारहवीं +१९वां~उन्नीसवां +२०वां~बीसवां +२१वां~इक्कीसवां +२५वीं~पच्चीसवीं +२७वें~सत्ताईसवें +३०वीं~तीसवीं +३३वां~तैंतीसवां +४०वीं~चालीसवीं +४५वां~पैंतालीसवां +५०वां~पचासवां +५६वें~छप्पनवें +६०वां~साठवां +६७वीं~सड़सठवीं +७५वीं~पचहत्तरवीं +८०वें~अस्सीवें +८८वां~अट्ठासीवां +९१वीं~इक्यानबेवीं +९९वां~निन्यानबेवां +१००वां~एक सौवां +१०१वां~एक सौ एकवां +१११वीं~एक सौ ग्यारहवीं +१२५वें~एक सौ पच्चीसवें +१५३वीं~एक सौ तिरेपनवीं +२००वीं~दो सौवीं +२१९वीं~दो सौ उन्नीसवीं +२४०वां~दो सौ चालीसवां +३२९वां~तीन सौ उनतीसवां +३६५वां~तीन सौ पैंसठवां +४५५वां~चार सौ पचपनवां +५५५वीं~पाँच सौ पचपनवीं +६४०वीं~छह सौ चालीसवीं +८९०वां~आठ सौ नब्बेवां +१००१वीं~एक हज़ार एकवीं +१०९१वें~एक हज़ार इक्यानबेवें +१७८२वीं~सत्रह सौ बयासीवीं +१८९०वां~एक हज़ार आठ सौ नब्बेवां +१९८१वीं~उन्नीस सौ इक्यासीवीं +९८२६वीं~अट्ठानबे सौ छब्बीसवीं \ No newline at end of file diff --git a/tests/nemo_text_processing/hi/data_text_normalization/test_cases_telephone.txt b/tests/nemo_text_processing/hi/data_text_normalization/test_cases_telephone.txt new file mode 100644 index 000000000..7a1b2c662 --- /dev/null +++ b/tests/nemo_text_processing/hi/data_text_normalization/test_cases_telephone.txt @@ -0,0 +1,25 @@ +मेरा पुराना नंबर था ९१५७११४००७~मेरा पुराना नंबर था शून्य नौ एक पाँच सात एक एक चार शून्य शून्य सात +इसपे कॉल करो ०३८६२-३५१७९१~इसपे कॉल करो शून्य तीन आठ छह दो तीन पाँच एक सात नौ एक +मेरे इस नंबर पे कॉल करो १३७४-३०९९८८~मेरे इस नंबर पे कॉल करो शून्य एक तीन सात चार तीन शून्य नौ नौ आठ आठ +इसपे कॉल करो ०१६८९११-४५७३~इसपे कॉल करो शून्य एक छह आठ नौ एक एक चार पाँच सात तीन ++९१ ७४४०४३१०८३ मेरे इस नंबर पे कॉल करो~प्लस नौ एक सात चार चार शून्य चार तीन एक शून्य आठ तीन मेरे इस नंबर पे कॉल करो ++९१ ९२१०५१५६०६ मेरे इस नंबर पे कॉल करो~प्लस नौ एक नौ दो एक शून्य पाँच एक पाँच छह शून्य छह मेरे इस नंबर पे कॉल करो +भुगतान के लिए कार्ड के आखिरी अंक १२३४ दर्ज करें~भुगतान के लिए कार्ड के आखिरी अंक एक दो तीन चार दर्ज करें +मेरा पिन कोड ११००२३ है~मेरा पिन कोड एक एक शून्य शून्य दो तीन है +मेरा पुराना नंबर था 9157114007~मेरा पुराना नंबर था शून्य नौ एक पाँच सात एक एक चार शून्य शून्य सात +इसपे कॉल करो 03862-351791~इसपे कॉल करो शून्य तीन आठ छह दो तीन पाँच एक सात नौ एक +मेरे इस नंबर पे कॉल करो 1374 309988~मेरे इस नंबर पे कॉल करो शून्य एक तीन सात चार तीन शून्य नौ नौ आठ आठ +इसपे कॉल करो 0168911-4573~इसपे कॉल करो शून्य एक छह आठ नौ एक एक चार पाँच सात तीन ++91 7440431083 मेरे इस नंबर पे कॉल करो~प्लस नौ एक सात चार चार शून्य चार तीन एक शून्य आठ तीन मेरे इस नंबर पे कॉल करो ++91 9210515606 मेरे इस नंबर पे कॉल करो~प्लस नौ एक नौ दो एक शून्य पाँच एक पाँच छह शून्य छह मेरे इस नंबर पे कॉल करो +भुगतान के लिए कार्ड के आखिरी अंक 1234 दर्ज करें~भुगतान के लिए कार्ड के आखिरी अंक एक दो तीन चार दर्ज करें +मेरा पिन कोड 110023 है~मेरा पिन कोड एक एक शून्य शून्य दो तीन है ++1 9210515606 मेरे इस नंबर पे कॉल करो~प्लस एक नौ दो एक शून्य पाँच एक पाँच छह शून्य छह मेरे इस नंबर पे कॉल करो ++४९ ९२१०५१५६०६ मेरे इस नंबर पे कॉल करो~प्लस चार नौ नौ दो एक शून्य पाँच एक पाँच छह शून्य छह मेरे इस नंबर पे कॉल करो ++353 9210515606 मेरे इस नंबर पे कॉल करो~प्लस तीन पाँच तीन नौ दो एक शून्य पाँच एक पाँच छह शून्य छह मेरे इस नंबर पे कॉल करो ++91 9876543210 123~प्लस नौ एक नौ आठ सात छह पाँच चार तीन दो एक शून्य एक दो तीन ++1 6234517890 123~प्लस एक छह दो तीन चार पाँच एक सात आठ नौ शून्य एक दो तीन ++९१ ९८७६५४३२१० १२३~प्लस नौ एक नौ आठ सात छह पाँच चार तीन दो एक शून्य एक दो तीन +(02229) 411128~शून्य दो दो दो नौ चार एक एक एक दो आठ +०२२.२९४१११२८~शून्य दो दो दो नौ चार एक एक एक दो आठ +0 (80) 26411128~शून्य आठ शून्य दो छह चार एक एक एक दो आठ \ No newline at end of file diff --git a/tests/nemo_text_processing/hi/test_ordinal.py b/tests/nemo_text_processing/hi/test_ordinal.py index b65252694..3e5f4bfbb 100644 --- a/tests/nemo_text_processing/hi/test_ordinal.py +++ b/tests/nemo_text_processing/hi/test_ordinal.py @@ -17,13 +17,24 @@ from parameterized import parameterized from nemo_text_processing.inverse_text_normalization.inverse_normalize import InverseNormalizer +from nemo_text_processing.text_normalization.normalize import Normalizer from ..utils import CACHE_DIR, parse_test_case_file class TestOrdinal: + normalizer = Normalizer( + input_case='cased', lang='hi', cache_dir=CACHE_DIR, overwrite_cache=False, post_process=False + ) inverse_normalizer = InverseNormalizer(lang='hi', cache_dir=CACHE_DIR, overwrite_cache=False) + @parameterized.expand(parse_test_case_file('hi/data_text_normalization/test_cases_ordinal.txt')) + @pytest.mark.run_only_on('CPU') + @pytest.mark.unit + def test_norm(self, test_input, expected): + pred = self.normalizer.normalize(test_input, verbose=False) + assert pred.strip() == expected.strip() + @parameterized.expand(parse_test_case_file('hi/data_inverse_text_normalization/test_cases_ordinal.txt')) @pytest.mark.run_only_on('CPU') @pytest.mark.unit diff --git a/tests/nemo_text_processing/hi/test_sparrowhawk_normalization.sh b/tests/nemo_text_processing/hi/test_sparrowhawk_normalization.sh index 498443f71..a0b0931e2 100644 --- a/tests/nemo_text_processing/hi/test_sparrowhawk_normalization.sh +++ b/tests/nemo_text_processing/hi/test_sparrowhawk_normalization.sh @@ -76,15 +76,15 @@ testTNMoney() { runtest $input } -#testTNOrdinal() { -# input=$PROJECT_DIR/hi/data_text_normalization/test_cases_ordinal.txt -# runtest $input -#} +testTNOrdinal() { + input=$PROJECT_DIR/hi/data_text_normalization/test_cases_ordinal.txt + runtest $input +} -#testTNTelephone() { -# input=$PROJECT_DIR/en/data_text_normalization/test_cases_telephone.txt -# runtest $input -#} +testTNTelephone() { + input=$PROJECT_DIR/hi/data_text_normalization/test_cases_telephone.txt + runtest $input +} testTNTime() { input=$PROJECT_DIR/hi/data_text_normalization/test_cases_time.txt diff --git a/tests/nemo_text_processing/hi/test_telephone.py b/tests/nemo_text_processing/hi/test_telephone.py index 7e43f7e82..e7b9f1c3d 100644 --- a/tests/nemo_text_processing/hi/test_telephone.py +++ b/tests/nemo_text_processing/hi/test_telephone.py @@ -16,12 +16,16 @@ from parameterized import parameterized from nemo_text_processing.inverse_text_normalization.inverse_normalize import InverseNormalizer +from nemo_text_processing.text_normalization.normalize import Normalizer from ..utils import CACHE_DIR, parse_test_case_file class TestTelephone: inverse_normalizer = InverseNormalizer(lang='hi', cache_dir=CACHE_DIR, overwrite_cache=False) + normalizer = Normalizer( + input_case='cased', lang='hi', cache_dir=CACHE_DIR, overwrite_cache=False, post_process=True + ) @parameterized.expand(parse_test_case_file('hi/data_inverse_text_normalization/test_cases_telephone.txt')) @pytest.mark.run_only_on('CPU') @@ -29,3 +33,10 @@ class TestTelephone: def test_denorm(self, test_input, expected): pred = self.inverse_normalizer.inverse_normalize(test_input, verbose=False) assert pred.strip() == expected.strip() + + @parameterized.expand(parse_test_case_file('hi/data_text_normalization/test_cases_telephone.txt')) + @pytest.mark.run_only_on('CPU') + @pytest.mark.unit + def test_norm(self, test_input, expected): + pred = self.normalizer.normalize(test_input, verbose=False, punct_post_process=True) + assert pred == expected