Added cardinal semiotic class for te ITN

Jenkinsfile

@@ -27,6 +27,7 @@ pipeline {
     HE_TN_CACHE='/home/jenkins/TestData/text_norm/ci/grammars/09-24-25-0'
     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'
+    TE_TN_CACHE='/home/jenkins/TestData/text_norm/ci/grammars/07-22-26-0'
     JA_TN_CACHE='/home/jenkins/TestData/text_norm/ci/grammars/10-17-24-1'
     HI_TN_CACHE='/home/jenkins/TestData/text_norm/ci/grammars/06-04-26-5'
     KO_TN_CACHE='/home/jenkins/TestData/text_norm/ci/grammars/06-04-25-6'
@@ -297,6 +298,11 @@ pipeline {
             sh 'CUDA_VISIBLE_DEVICES="" python nemo_text_processing/inverse_text_normalization/inverse_normalize.py --lang=mr --text="शून्य " --cache_dir ${MR_TN_CACHE}'
           }
         }
+        stage('L0: TE ITN grammars') {
+          steps {
+            sh 'CUDA_VISIBLE_DEVICES="" python nemo_text_processing/inverse_text_normalization/inverse_normalize.py --lang=te --text="సున్నా " --cache_dir ${TE_TN_CACHE}'
+          }
+        }
         stage('L0: HY TN grammars') {
           steps {
             sh 'CUDA_VISIBLE_DEVICES="" python nemo_text_processing/text_normalization/normalize.py --lang=hy --text="6" --cache_dir ${HY_TN_CACHE}'
@@ -463,6 +469,11 @@ pipeline {
             sh 'CUDA_VISIBLE_DEVICES="" pytest tests/nemo_text_processing/mr/ -m "not pleasefixme" --cpu --tn_cache_dir ${MR_TN_CACHE}'
           }
         }
+        stage('L1: Run all TE ITN tests (restore grammars from cache)') {
+          steps {
+            sh 'CUDA_VISIBLE_DEVICES="" pytest tests/nemo_text_processing/te/ -m "not pleasefixme" --cpu --tn_cache_dir ${TE_TN_CACHE}'
+          }
+        }
         stage('L1: Run all HY TN/ITN tests (restore grammars from cache)') {
           steps {
             sh 'CUDA_VISIBLE_DEVICES="" pytest tests/nemo_text_processing/hy/ -m "not pleasefixme" --cpu --tn_cache_dir ${HY_TN_CACHE}'

nemo_text_processing/inverse_text_normalization/inverse_normalize.py

@@ -121,6 +121,11 @@ def __init__(
             from nemo_text_processing.inverse_text_normalization.hi.verbalizers.verbalize_final import (
                 VerbalizeFinalFst,
             )
+        elif lang == 'te':  # Telugu
+            from nemo_text_processing.inverse_text_normalization.te.taggers.tokenize_and_classify import ClassifyFst
+            from nemo_text_processing.inverse_text_normalization.te.verbalizers.verbalize_final import (
+                VerbalizeFinalFst,
+            )
         elif lang == 'hi_en':  # Hindi-English code-switch
             from nemo_text_processing.inverse_text_normalization.hi_en.taggers.tokenize_and_classify import ClassifyFst
             from nemo_text_processing.inverse_text_normalization.hi_en.verbalizers.verbalize_final import (
@@ -207,6 +212,7 @@ def parse_args():
             'zh',
             'he',
             'hi',
+            'te',
             'hy',
             'mr',
             'ja',

nemo_text_processing/inverse_text_normalization/run_evaluate.py

@@ -44,6 +44,7 @@ def parse_args():
             "fr",
             "hi",
             "hi_en",
+            "te",
             "hy",
             "ko",
             "mr",

nemo_text_processing/inverse_text_normalization/te/__init__.py

@@ -0,0 +1,17 @@
+# Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. 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.
+
+from nemo_text_processing.inverse_text_normalization.te.taggers.tokenize_and_classify import ClassifyFst
+from nemo_text_processing.inverse_text_normalization.te.verbalizers.verbalize import VerbalizeFst
+from nemo_text_processing.inverse_text_normalization.te.verbalizers.verbalize_final import VerbalizeFinalFst

nemo_text_processing/inverse_text_normalization/te/data/__init__.py

@@ -0,0 +1,13 @@
+# Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. 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.

nemo_text_processing/inverse_text_normalization/te/data/numbers/__init__.py

@@ -0,0 +1,13 @@
+# Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. 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.

nemo_text_processing/inverse_text_normalization/te/data/numbers/case_suffix.tsv

@@ -0,0 +1,5 @@
+కి
+కు
+లో
+తో
+కోసం

nemo_text_processing/inverse_text_normalization/te/data/numbers/crore.tsv

@@ -0,0 +1,3 @@
+కోట్ల	
+కోట్లు	
+కోటి	౦౦౦౦౦౦౧

nemo_text_processing/inverse_text_normalization/te/data/numbers/digit.tsv

@@ -0,0 +1,9 @@
+ఒకటి	౧
+రెండు	౨
+మూడు	౩
+నాలుగు	౪
+ఐదు	౫
+ఆరు	౬
+ఏడు	౭
+ఎనిమిది	౮
+తొమ్మిది	౯

nemo_text_processing/inverse_text_normalization/te/data/numbers/digit_people.tsv

@@ -0,0 +1,9 @@
+ఇద్దరు	౨
+ముగ్గురు	౩
+నలుగురు	౪
+ఐదుగురు	౫
+ఆరుగురు	౬
+ఏడుగురు	౭
+ఎనిమిదుగురు	౮
+తొమ్మిదుగురు	౯
+పది మంది	౧౦

nemo_text_processing/inverse_text_normalization/te/data/numbers/hundred.tsv

@@ -0,0 +1,5 @@
+వందల	
+వందలు	
+నూట	౧
+వంద	౧
+నూరు	౧

nemo_text_processing/inverse_text_normalization/te/data/numbers/lakh.tsv

@@ -0,0 +1,4 @@
+లక్షల	
+లక్షలు	
+లక్ష	౦౧
+లక్షా	౦౧

nemo_text_processing/inverse_text_normalization/te/data/numbers/special_numbers.tsv

@@ -0,0 +1,2 @@
+పంతొమ్మిది	౧౯
+ఒక	౧

nemo_text_processing/inverse_text_normalization/te/data/numbers/teens_and_ties.tsv

@@ -0,0 +1,18 @@
+పది	౧౦
+పదకొండు	౧౧
+పన్నెండు	౧౨
+పదమూడు	౧౩
+పద్నాలుగు	౧౪
+పదిహేను	౧౫
+పదహారు	౧౬
+పదిహేడు	౧౭
+పద్దెనిమిది	౧౮
+పందొమ్మిది	౧౯
+ఇరవై	౨౦
+ముప్పై	౩౦
+నలభై	౪౦
+యాభై	౫౦
+అరవై	౬౦
+డెబ్బై	౭౦
+ఎనభై	౮౦
+తొంభై	౯౦

nemo_text_processing/inverse_text_normalization/te/data/numbers/thousand.tsv

@@ -0,0 +1,3 @@
+వేల	
+వేలు	
+వెయ్యి	౦౧

nemo_text_processing/inverse_text_normalization/te/data/numbers/ties_prefix.tsv

@@ -0,0 +1,8 @@
+ఇరవై	౨
+ముప్పై	౩
+నలభై	౪
+యాభై	౫
+అరవై	౬
+డెబ్బై	౭
+ఎనభై	౮
+తొంభై	౯

nemo_text_processing/inverse_text_normalization/te/data/numbers/zero.tsv

@@ -0,0 +1 @@
+సున్నా	౦

nemo_text_processing/inverse_text_normalization/te/graph_utils.py

@@ -0,0 +1,195 @@
+# Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. 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 logging
+import os
+import string
+from pathlib import Path
+from typing import Dict
+
+import pynini
+from pynini import Far
+from pynini.export import export
+from pynini.lib import pynutil, utf8
+
+from nemo_text_processing.inverse_text_normalization.te.utils import load_labels
+
+NEMO_CHAR = utf8.VALID_UTF8_CHAR
+
+NEMO_TE_DIGIT = pynini.union("౦", "౧", "౨", "౩", "౪", "౫", "౬", "౭", "౮", "౯").optimize()
+
+NEMO_HEX = pynini.union(*string.hexdigits).optimize()
+NEMO_NON_BREAKING_SPACE = u"\u00a0"
+NEMO_ZWNJ = u"\u200c"
+NEMO_SPACE = " "
+NEMO_WHITE_SPACE = pynini.union(" ", "\t", "\n", "\r", u"\u00a0").optimize()
+NEMO_NOT_SPACE = pynini.difference(NEMO_CHAR, NEMO_WHITE_SPACE).optimize()
+NEMO_NOT_QUOTE = pynini.difference(NEMO_CHAR, r'"').optimize()
+
+NEMO_PUNCT = pynini.union(*map(pynini.escape, string.punctuation)).optimize()
+NEMO_GRAPH = pynini.union(NEMO_CHAR, NEMO_PUNCT).optimize()
+
+NEMO_SIGMA = pynini.closure(NEMO_CHAR)
+
+delete_space = pynutil.delete(pynini.closure(NEMO_WHITE_SPACE))
+delete_zero_or_one_space = pynutil.delete(pynini.closure(NEMO_WHITE_SPACE, 0, 1))
+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("\""))
+)
+
+
+MIN_NEG_WEIGHT = -0.0001
+MIN_POS_WEIGHT = 0.0001
+INPUT_CASED = "cased"
+INPUT_LOWER_CASED = "lower_cased"
+MINUS = pynini.union("ఋణాత్మక", "మైనస్").optimize()
+
+
+def generator_main(file_name: str, graphs: Dict[str, 'pynini.FstLike']):
+    """
+    Exports graph as OpenFst finite state archive (FAR) file with given file name and rule name.
+
+    Args:
+        file_name: exported file name
+        graphs: Mapping of a rule name and Pynini WFST graph to be exported
+    """
+    exporter = export.Exporter(file_name)
+    for rule, graph in graphs.items():
+        exporter[rule] = graph.optimize()
+    exporter.close()
+    logging.info(f'Created {file_name}')
+
+
+def convert_space(fst) -> 'pynini.FstLike':
+    """
+    Converts space to nonbreaking space.
+    Used only in tagger grammars for transducing token values within quotes, e.g. name: "hello kitty"
+    This is making transducer significantly slower, so only use when there could be potential spaces within quotes, otherwise leave it.
+
+    Args:
+        fst: input fst
+
+    Returns output fst where breaking spaces are converted to non breaking spaces
+    """
+    return fst @ pynini.cdrewrite(pynini.cross(NEMO_SPACE, NEMO_NON_BREAKING_SPACE), "", "", NEMO_SIGMA)
+
+
+def string_map_cased(input_file: str, input_case: str = INPUT_LOWER_CASED):
+    labels = load_labels(input_file)
+
+    if input_case == INPUT_CASED:
+        additional_labels = []
+        for written, spoken, *weight in labels:
+            written_capitalized = written[0].upper() + written[1:]
+            additional_labels.extend(
+                [
+                    [written_capitalized, spoken.capitalize()],
+                    [
+                        written_capitalized,
+                        spoken.upper().replace(" AND ", " and "),
+                    ],
+                ]
+            )
+
+            spoken_no_space = spoken.replace(" ", "")
+
+            if len(spoken) == (2 * len(spoken_no_space) - 1):
+                logging.debug(f"This is weight {weight}")
+                if len(weight) == 0:
+                    additional_labels.extend(
+                        [[written, spoken_no_space], [written_capitalized, spoken_no_space.upper()]]
+                    )
+                else:
+                    additional_labels.extend(
+                        [
+                            [written, spoken_no_space, weight[0]],
+                            [written_capitalized, spoken_no_space.upper(), weight[0]],
+                        ]
+                    )
+        labels += additional_labels
+
+    whitelist = pynini.string_map(labels).invert().optimize()
+    return whitelist
+
+
+class GraphFst:
+    """
+    Base class for all grammar fsts.
+
+    Args:
+        name: name of grammar class
+        kind: either 'classify' or 'verbalize'
+        deterministic: if True will provide a single transduction option,
+            for False multiple transduction are generated (used for audio-based normalization)
+    """
+
+    def __init__(self, name: str, kind: str, deterministic: bool = True):
+        self.name = name
+        self.kind = kind
+        self._fst = None
+        self.deterministic = deterministic
+
+        self.far_path = Path(os.path.dirname(__file__) + '/grammars/' + kind + '/' + name + '.far')
+        if self.far_exist():
+            self._fst = Far(self.far_path, mode="r", arc_type="standard", far_type="default").get_fst()
+
+    def far_exist(self) -> bool:
+        """
+        Returns true if FAR can be loaded
+        """
+        return self.far_path.exists()
+
+    @property
+    def fst(self) -> 'pynini.FstLike':
+        return self._fst
+
+    @fst.setter
+    def fst(self, fst):
+        self._fst = fst
+
+    def add_tokens(self, fst) -> 'pynini.FstLike':
+        """
+        Wraps class name around to given fst
+
+        Args:
+            fst: input fst
+
+        Returns:
+            Fst: fst
+        """
+        return pynutil.insert(f"{self.name} {{ ") + fst + pynutil.insert(" }")
+
+    def delete_tokens(self, fst) -> 'pynini.FstLike':
+        """
+        Deletes class name wrap around output of given fst
+
+        Args:
+            fst: input fst
+
+        Returns:
+            Fst: fst
+        """
+        res = (
+            pynutil.delete(f"{self.name}")
+            + delete_space
+            + pynutil.delete("{")
+            + delete_space
+            + fst
+            + delete_space
+            + pynutil.delete("}")
+        )
+        return res @ pynini.cdrewrite(pynini.cross(u"\u00a0", " "), "", "", NEMO_SIGMA)

nemo_text_processing/inverse_text_normalization/te/taggers/__init__.py

@@ -0,0 +1,13 @@
+# Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. 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.