spanish tn code switch

nemo_text_processing/text_normalization/en/data/address/address_word.tsv

@@ -4,6 +4,8 @@ expy	Expressway
 fwy	Freeway
 hwy	Highway
 dr	Drive
+rd	Road
+road	Road
 ct	Court
 ave	Avenue
 av	Avenue

nemo_text_processing/text_normalization/es/data/address/__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/text_normalization/es/data/address/apt_designator.tsv

@@ -0,0 +1,6 @@
+Apt.	Apartamento
+Apt.	Apartamento
+Dept.	Departamento
+Dept	Departamento
+Depto.	Departamento
+Depto	Departamento

nemo_text_processing/text_normalization/es/data/address/direction.tsv

@@ -0,0 +1,4 @@
+E	East
+S	South
+W	West
+N	North

nemo_text_processing/text_normalization/es/data/address/po_box.tsv

@@ -0,0 +1 @@
+P.O. Box 	P.O. Box 

nemo_text_processing/text_normalization/es/data/address/suite_designator.tsv

@@ -0,0 +1,2 @@
+Ste.	Suite 
+Ste 	Suite 

nemo_text_processing/text_normalization/es/data/address/unit_designator.tsv

@@ -0,0 +1 @@
+Unit 	Unit 

nemo_text_processing/text_normalization/es/graph_utils.py

@@ -133,6 +133,21 @@ def strip_cardinal_apocope(fst: "pynini.FstLike") -> "pynini.FstLike":
     return fst @ strip
 
 
+def normalize_spanish_cardinal_for_us_address_street(fst: "pynini.FstLike") -> "pynini.FstLike":
+    """
+    Spanish cardinals often apocopate before a following vowel (e.g. ``veintiún``). US street names
+    are ASCII and usually start with a consonant, but the cardinal FST does not see that context when
+    materializing digits alone. Normalize common ``…ún`` spoken forms to ``…uno`` / ``… y uno`` for
+    address surfaces (same intent as ``strip_cardinal_apocope`` but not restricted to string end).
+    """
+    out = fst
+    out = out @ pynini.cdrewrite(pynini.cross("veintiún", "veintiuno"), "", "", NEMO_SIGMA)
+    out = out @ pynini.cdrewrite(pynini.cross("treintún", "treinta y uno"), "", "", NEMO_SIGMA)
+    out = out @ pynini.cdrewrite(pynini.cross(" y ún", " y uno"), "", "", NEMO_SIGMA)
+    out = out @ pynini.cdrewrite(pynini.cross(" y un", " y uno"), "", "", NEMO_SIGMA)
+    return strip_cardinal_apocope(out)
+
+
 def add_cardinal_apocope_fem(fst: "pynini.FstLike") -> "pynini.FstLike":
     """
     Adds apocope on cardinal strings in line with stressing rules. e.g. "una" -> "un". This only occurs when "una" precedes a stressed "a" sound in formal speech. This is not predictable

nemo_text_processing/text_normalization/es/taggers/address.py

@@ -0,0 +1,157 @@
+# 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.
+
+"""
+US-style postal address surface for Spanish TN (embedded in ``MeasureFst`` as
+``units: "address_us_es"``).
+
+Street numbers and ZIP are Spanish; street types, states, and ordinals (e.g. ``42nd``)
+use English expansions from shared ``en/data/address/`` lexicons.
+"""
+
+import pynini
+from pynini.examples import plurals
+from pynini.lib import pynutil
+
+from nemo_text_processing.text_normalization.en.graph_utils import (
+    NEMO_ALPHA,
+    NEMO_DIGIT,
+    NEMO_SIGMA,
+    NEMO_SPACE,
+    NEMO_UPPER,
+    GraphFst,
+    insert_space,
+)
+from nemo_text_processing.text_normalization.en.taggers.cardinal import CardinalFst as EnCardinalFst
+from nemo_text_processing.text_normalization.en.taggers.ordinal import OrdinalFst as OrdinalTagger
+from nemo_text_processing.text_normalization.en.taggers.whitelist import get_formats
+from nemo_text_processing.text_normalization.en.utils import get_abs_path as en_get_abs_path
+from nemo_text_processing.text_normalization.en.utils import load_labels
+from nemo_text_processing.text_normalization.en.verbalizers.ordinal import OrdinalFst as OrdinalVerbalizer
+from nemo_text_processing.text_normalization.es.graph_utils import normalize_spanish_cardinal_for_us_address_street
+from nemo_text_processing.text_normalization.es.utils import get_abs_path
+
+
+class AddressUSSurfaceFst(GraphFst):
+    """
+    Surface FST for US addresses inside Spanish sentences.
+
+    Output is the spoken string stored in ``measure { units: "address_us_es" cardinal { integer: "..." } }``.
+    Not registered in ``tokenize_and_classify``; consumed by :class:`~nemo_text_processing.text_normalization.es.taggers.measure.MeasureFst`.
+
+    Args:
+        cardinal: Spanish :class:`~nemo_text_processing.text_normalization.es.taggers.cardinal.CardinalFst`
+        deterministic: passed to English ordinal/cardinal helpers
+    """
+
+    def __init__(self, cardinal: GraphFst, deterministic: bool = True):
+        super().__init__(name="address_us_es_surface", kind="classify", deterministic=deterministic)
+
+        graph_direction = pynini.string_file(get_abs_path("data/address/direction.tsv"))
+        graph_zip_digit = pynini.invert(
+            pynini.string_file(get_abs_path("data/numbers/zero.tsv"))
+            | pynini.string_file(get_abs_path("data/numbers/digit.tsv"))
+        ).optimize()
+        graph_zip_digit @= pynini.cdrewrite(pynini.cross("un", "uno"), "", "", NEMO_SIGMA)
+        graph_suite_designator = pynini.string_file(get_abs_path("data/address/suite_designator.tsv"))
+        graph_apt_designator = pynini.string_file(get_abs_path("data/address/apt_designator.tsv"))
+        graph_unit_designator = pynini.string_file(get_abs_path("data/address/unit_designator.tsv"))
+        graph_po_box = pynini.string_file(get_abs_path("data/address/po_box.tsv"))
+
+        en_cardinal = EnCardinalFst(deterministic=deterministic)
+        g = cardinal.graph
+
+        ordinal_en = pynini.compose(
+            pynutil.insert('integer: "') + OrdinalTagger(cardinal=en_cardinal).graph + pynutil.insert('"'),
+            OrdinalVerbalizer().graph,
+        )
+
+        address_num = NEMO_DIGIT ** (1, 2) @ cardinal.graph_hundreds_component_at_least_one_none_zero_digit
+        address_num += insert_space + NEMO_DIGIT**2 @ (
+            pynini.closure(pynini.cross("0", "cero "), 0, 1)
+            + cardinal.graph_hundreds_component_at_least_one_none_zero_digit
+        )
+        address_num = pynini.compose(NEMO_DIGIT ** (3, 4), address_num)
+        address_num = normalize_spanish_cardinal_for_us_address_street(
+            plurals._priority_union(address_num, g, NEMO_SIGMA).optimize()
+        )
+
+        direction = pynini.closure(
+            pynini.accep(NEMO_SPACE) + graph_direction + pynini.closure(pynutil.delete("."), 0, 1),
+            0,
+            1,
+        )
+
+        address_words = get_formats(en_get_abs_path("data/address/address_word.tsv"))
+        street = (
+            pynini.accep(NEMO_SPACE)
+            + (pynini.closure(ordinal_en, 0, 1) | NEMO_UPPER + pynini.closure(NEMO_ALPHA, 1))
+            + NEMO_SPACE
+            + pynini.closure(NEMO_UPPER + pynini.closure(NEMO_ALPHA) + NEMO_SPACE)
+            + address_words
+        )
+
+        zip_five = (pynini.closure(graph_zip_digit + insert_space, 4) + graph_zip_digit).optimize()
+
+        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(en_get_abs_path("data/address/state.tsv"))
+        states_extra = [(x, f"{y[0]}.{y[1:]}") for x, y in states]
+        states.extend(states_extra)
+        state = pynini.closure(
+            pynini.accep(",") + pynini.accep(NEMO_SPACE) + pynini.invert(pynini.string_map(states)), 0, 1
+        )
+
+        zip_code = pynini.closure(
+            pynini.closure(pynini.accep(","), 0, 1) + pynini.accep(NEMO_SPACE) + zip_five,
+            0,
+            1,
+        )
+        tail = pynini.closure(city + state + zip_code, 0, 1).optimize()
+
+        suite_num = normalize_spanish_cardinal_for_us_address_street((pynini.closure(NEMO_DIGIT, 1, 4) @ g).optimize())
+        unit_num = normalize_spanish_cardinal_for_us_address_street((pynini.closure(NEMO_DIGIT, 1, 3) @ g).optimize())
+        apt_char = graph_zip_digit | NEMO_UPPER
+        apt_num = (apt_char + pynini.closure(insert_space + apt_char, 0, 3)).optimize()
+
+        comma_sp = pynini.accep(",") + pynini.accep(NEMO_SPACE)
+        suite = graph_suite_designator + pynini.closure(NEMO_SPACE, 0, 1) + suite_num
+        apt = graph_apt_designator + pynini.closure(NEMO_SPACE, 0, 1) + apt_num
+        unit = graph_unit_designator + unit_num
+        middle = pynini.closure(comma_sp + (suite | apt | unit), 0, 3).optimize()
+
+        po_box = (
+            graph_po_box
+            + normalize_spanish_cardinal_for_us_address_street(pynini.closure(NEMO_DIGIT, 1, 4) @ g)
+            + tail
+        ).optimize()
+
+        standard = address_num + direction + street + middle + tail
+        hyphen = pynini.accep("-")
+        alpha_chars = NEMO_ALPHA | hyphen
+        standard_eos = (
+            address_num
+            + direction
+            + street
+            + middle
+            + pynini.accep(".")
+            + pynini.closure(NEMO_SPACE, 1, 2)
+            + NEMO_UPPER
+            + pynini.closure(alpha_chars)
+        )
+        standard |= pynutil.add_weight(standard_eos, -0.001)
+        standard |= address_num + direction + street + middle + pynini.closure(pynini.cross(".", ""), 0, 1)
+
+        self.graph = (po_box | standard.optimize()).optimize()

nemo_text_processing/text_normalization/es/taggers/cardinal.py

@@ -169,6 +169,10 @@ def __init__(self, deterministic: bool = True):
 
         self.graph = filter_punctuation(self.graph).optimize()
 
+        self.graph_hundreds_component_at_least_one_none_zero_digit = (
+            graph_hundreds_component_at_least_one_none_zero_digit.optimize()
+        )
+
         optional_minus_graph = pynini.closure(pynutil.insert("negative: ") + pynini.cross("-", "\"true\" "), 0, 1)
 
         final_graph = optional_minus_graph + pynutil.insert("integer: \"") + self.graph + pynutil.insert("\"")

nemo_text_processing/text_normalization/es/taggers/measure.py

@@ -25,6 +25,7 @@
     insert_space,
 )
 from nemo_text_processing.text_normalization.es.graph_utils import strip_cardinal_apocope
+from nemo_text_processing.text_normalization.es.taggers.address import AddressUSSurfaceFst
 from nemo_text_processing.text_normalization.es.utils import get_abs_path
 
 unit = pynini.string_file(get_abs_path("data/measures/measurements.tsv"))
@@ -199,6 +200,13 @@ def __init__(self, cardinal: GraphFst, decimal: GraphFst, fraction: GraphFst, de
             + pynutil.insert("\" } preserve_order: true")
         )
 
+        address_us_es_inner = AddressUSSurfaceFst(cardinal, deterministic=deterministic).graph
+        address_us_es = (
+            pynutil.insert('units: "address_us_es" cardinal { integer: "')
+            + address_us_es_inner
+            + pynutil.insert('" } preserve_order: true')
+        )
+
         final_graph = (
             subgraph_decimal
             | subgraph_cardinal
@@ -210,6 +218,7 @@ def __init__(self, cardinal: GraphFst, decimal: GraphFst, fraction: GraphFst, de
             | cardinal_times
             | alpha_dash_decimal
             | math
+            | address_us_es
         )
         final_graph = self.add_tokens(final_graph)
 

nemo_text_processing/text_normalization/es/verbalizers/measure.py

@@ -23,6 +23,7 @@
     delete_extra_space,
     delete_preserve_order,
     delete_space,
+    insert_space,
 )
 from nemo_text_processing.text_normalization.es.graph_utils import ones
 from nemo_text_processing.text_normalization.es.utils import get_abs_path
@@ -65,12 +66,20 @@ def __init__(self, decimal: GraphFst, cardinal: GraphFst, fraction: GraphFst, de
             NEMO_WHITE_SPACE + "por" + pynini.closure(NEMO_NOT_QUOTE, 1), 0, 1
         )
         unit_masc |= "por" + pynini.closure(NEMO_NOT_QUOTE, 1)
-        unit_masc = pynutil.delete("units: \"") + (pynini.closure(NEMO_NOT_QUOTE) @ unit_masc) + pynutil.delete("\"")
+        unit_masc = (
+            pynutil.delete("units: \"")
+            + (pynini.difference(pynini.closure(NEMO_NOT_QUOTE, 1), pynini.union("math", "address_us_es")) @ unit_masc)
+            + pynutil.delete("\"")
+        )
 
         unit_fem = (unit_plural_fem | unit_singular_fem) + pynini.closure(
             NEMO_WHITE_SPACE + "por" + pynini.closure(NEMO_NOT_QUOTE, 1), 0, 1
         )
-        unit_fem = pynutil.delete("units: \"") + (pynini.closure(NEMO_NOT_QUOTE) @ unit_fem) + pynutil.delete("\"")
+        unit_fem = (
+            pynutil.delete("units: \"")
+            + (pynini.difference(pynini.closure(NEMO_NOT_QUOTE, 1), pynini.union("math", "address_us_es")) @ unit_fem)
+            + pynutil.delete("\"")
+        )
 
         graph_masc = (graph_cardinal_masc | graph_decimal_masc) + NEMO_WHITE_SPACE + unit_masc
         graph_masc |= graph_fraction_masc + NEMO_WHITE_SPACE + pynutil.insert("de ") + unit_masc
@@ -96,7 +105,11 @@ def __init__(self, decimal: GraphFst, cardinal: GraphFst, fraction: GraphFst, de
         graph @= pynini.cdrewrite(pynini.cross(ones, "uno"), "", NEMO_WHITE_SPACE + "por", NEMO_SIGMA)
 
         # To manage alphanumeric combonations ("a-8, 5x"), we let them use a weighted default path.
-        alpha_num_unit = pynutil.delete("units: \"") + pynini.closure(NEMO_NOT_QUOTE) + pynutil.delete("\"")
+        alpha_num_unit = (
+            pynutil.delete("units: \"")
+            + pynini.difference(pynini.closure(NEMO_NOT_QUOTE), pynini.union("math", "address_us_es"))
+            + pynutil.delete("\"")
+        )
         graph_alpha_num = pynini.union(
             (graph_cardinal_masc | graph_decimal_masc) + NEMO_SPACE + alpha_num_unit,
             alpha_num_unit + delete_extra_space + (graph_cardinal_masc | graph_decimal_masc),
@@ -106,8 +119,23 @@ def __init__(self, decimal: GraphFst, cardinal: GraphFst, fraction: GraphFst, de
             pynutil.delete("units: \"math\"") + delete_space + graph_cardinal_masc + delete_space, -1
         )
 
+        preserve_order_tail = pynutil.delete("preserve_order:") + delete_space + pynutil.delete("true") + delete_space
+        address_us_es = (
+            pynutil.delete('units: "address_us_es" ')
+            + delete_space
+            + pynutil.delete("cardinal { integer: \"")
+            + delete_space
+            + pynini.closure(NEMO_NOT_QUOTE)
+            + pynutil.delete("\"")
+            + delete_space
+            + pynutil.delete("}")
+            + delete_space
+            + pynini.closure(preserve_order_tail)
+        )
+
         graph |= pynutil.add_weight(graph_alpha_num, 0.01)
         graph |= math
+        graph |= address_us_es
 
         graph += delete_preserve_order
 

tests/nemo_text_processing/es/data_text_normalization/test_cases_address.txt

@@ -0,0 +1,10 @@
+Mi dirección es 1234 Maple St., Springfield, IL 62704~Mi dirección es mil doscientos treinta y cuatro Maple Street, Springfield, Illinois seis dos siete cero cuatro
+La oficina está ubicada en 567 Main St., Ste. 200, Dallas, TX 75201~La oficina está ubicada en quinientos sesenta y siete Main Street, Suite doscientos, Dallas, Texas siete cinco dos cero uno
+Por favor envía el paquete a 890 Oak Ave., Apt. 5B, Brooklyn, NY 11201~Por favor envía el paquete a ochocientos noventa Oak Avenue, Apartamento cinco B, Brooklyn, New York uno uno dos cero uno
+Vivo en 4321 Sunset Blvd., Los Angeles, CA 90028, cerca del centro~Vivo en cuatro mil trescientos veintiuno Sunset Boulevard, Los Angeles, California nueve cero cero dos ocho, cerca del centro
+La nueva tienda abrirá en 100 Market Rd., San Francisco, CA 94105~La nueva tienda abrirá en cien Market Road, San Francisco, California nueve cuatro uno cero cinco
+Su casa queda en 25 W 42nd St., New York, NY 10036~Su casa queda en veinticinco West forty second Street, New York, New York uno cero cero tres seis
+El hospital se encuentra en 7890 Lincoln Dr., Miami, FL 33133~El hospital se encuentra en siete mil ochocientos noventa Lincoln Drive, Miami, Florida tres tres uno tres tres
+Mándame la carta a P.O. Box 456, Austin, TX 78701, por favor~Mándame la carta a P.O. Box cuatrocientos cincuenta y seis, Austin, Texas siete ocho siete cero uno, por favor
+La escuela de mis hijos está en 321 Elm St., Boston, MA 02108~La escuela de mis hijos está en trescientos veintiuno Elm Street, Boston, Massachusetts cero dos uno cero ocho
+Nos mudamos a 9876 Pine Rd., Unit 12, Seattle, WA 98101 el mes pasado~Nos mudamos a nueve mil ochocientos setenta y seis Pine Road, Unit doce, Seattle, Washington nueve ocho uno cero uno el mes pasado

tests/nemo_text_processing/es/data_text_normalization/test_cases_measure.txt

@@ -29,4 +29,4 @@ a-5~a cinco
 -8° c~menos ocho grados centígrados
 40 ° k~cuarenta grados kelvin
 180 psi~ciento ochenta p s i
-2 + 2 - 1 = 3~dos más dos menos uno es igual a tres
+2 + 2 - 1 = 3~dos más dos menos uno es igual a tres

tests/nemo_text_processing/es/test_address.py

@@ -0,0 +1,47 @@
+# 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 pytest
+from parameterized import parameterized
+
+from nemo_text_processing.text_normalization.normalize import Normalizer
+from nemo_text_processing.text_normalization.normalize_with_audio import NormalizerWithAudio
+
+from ..utils import CACHE_DIR, RUN_AUDIO_BASED_TESTS, parse_test_case_file
+
+
+class TestAddress:
+    normalizer_es = Normalizer(input_case='cased', lang='es', cache_dir=CACHE_DIR, overwrite_cache=False)
+
+    normalizer_with_audio_es = (
+        NormalizerWithAudio(input_case='cased', lang='es', cache_dir=CACHE_DIR, overwrite_cache=False)
+        if RUN_AUDIO_BASED_TESTS
+        else None
+    )
+
+    # Spanish US-address code-switching is tagged by the measure class.
+    @parameterized.expand(parse_test_case_file('es/data_text_normalization/test_cases_address.txt'))
+    @pytest.mark.run_only_on('CPU')
+    @pytest.mark.unit
+    def test_norm(self, test_input, expected):
+        pred = self.normalizer_es.normalize(test_input, verbose=False, punct_post_process=True)
+        assert pred == expected
+
+        if self.normalizer_with_audio_es:
+            pred_non_deterministic = self.normalizer_with_audio_es.normalize(
+                test_input,
+                n_tagged=500,
+                punct_post_process=False,
+            )
+            assert expected in pred_non_deterministic