scriptconv 0.0.4a7__tar.gz → 0.0.4a8__tar.gz

{scriptconv-0.0.4a7/scriptconv.egg-info → scriptconv-0.0.4a8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: scriptconv
-Version: 0.0.4a7
+Version: 0.0.4a8
 Summary: Zero-dependency script & phoneme-notation core — ISO-15924 detection & metadata, IPA↔ARPABET/X-SAMPA/Lexique/Kirshenbaum/Cotovía/RFE, Buckwalter↔Arabic, Hangul→jamo, kana
 License: Apache-2.0
 Project-URL: Homepage, https://github.com/TigreGotico/scriptconv

{scriptconv-0.0.4a7 → scriptconv-0.0.4a8}/scriptconv/diacritics.py

@@ -245,24 +245,27 @@ def _sense_diacritics_pt(text: str) -> str:
     return add_extra_diacritics(text)
 
 
-def diacritize(text: str, lang: str = "und", model=None,
-               phonikud_model=None, diacritizer_model=None, **_) -> str:
+def diacritize(text: str, lang: str = "und", diacritizer_model=None,
+               **kwargs) -> str:
     """Add pronunciation-disambiguating diacritics to *text* for *lang*.
 
-    Four backends, each restoring information ordinary orthography omits but
-    downstream G2P needs:
-
-    - Hebrew (``he``) — niqqud via phonikud. ``phonikud_model=`` is optional:
-      omitted, the small public phonikud ONNX model is auto-downloaded and
-      cached (``$SCRIPTCONV_CACHE``/``$XDG_CACHE_HOME``); pass a path or
-      zero-arg callable to override.
-    - Arabic (``ar``) — tashkeel via text2tashkeel (``[tashkeel]``,
-      ``model=``/``diacritizer_model=``).
+    ``diacritizer_model=`` is the one model knob for the diacritizer edge — the
+    model for whichever backend the language routes to (there is one such knob,
+    parallel to ``phonemizer_model=`` on the ``text -> ipa`` phonemizer edge):
+
+    - Hebrew (``he``) — niqqud via phonikud; ``diacritizer_model`` is the
+      phonikud ONNX path (or a zero-arg callable). Optional: when omitted the
+      small public phonikud model is auto-downloaded and cached
+      (``$SCRIPTCONV_CACHE``/``$XDG_CACHE_HOME``).
+    - Arabic (``ar``) — tashkeel via text2tashkeel (``[tashkeel]``);
+      ``diacritizer_model`` is the text2tashkeel model name (defaults to
+      ``rawi-ensemble``).
     - East Slavic, Bulgarian/Macedonian/Slovene, Latvian, Armenian, Georgian,
       and Turkic/Caucasian languages (``STRESS_LANGS``, 26 stressonnx tags) —
-      word stress via stressonnx (``[stress]``); stress is unwritten or
-      under-marked in these languages, and in East Slavic unstressed vowels
-      also reduce, so a missing mark can corrupt more than prosody.
+      word stress via stressonnx (``[stress]``); ``diacritizer_model`` is the
+      stressonnx model. Stress is unwritten or under-marked in these languages,
+      and in East Slavic unstressed vowels also reduce, so a missing mark can
+      corrupt more than prosody.
     - European Portuguese (``pt``/``pt-PT``, never ``pt-BR``) —
       heterophonic-homograph sense diacritics via bifonia (``[pt]``);
       ordinary Portuguese orthographic marks that any downstream G2P reads
@@ -272,13 +275,17 @@ def diacritize(text: str, lang: str = "und", model=None,
     ``ImportError`` naming its extra when the optional dependency is missing
     — scriptconv never installs anything on the caller's behalf.
     """
+    # one knob for the diacritizer edge; fold the pre-unification kwargs so
+    # older callers keep working (phonikud_model was the Hebrew-only name).
+    diacritizer_model = (diacritizer_model or kwargs.get("phonikud_model")
+                         or kwargs.get("model"))
     family = _diacritizer_family(lang)
     if family == "he":
-        return _phonikud(phonikud_model).add_diacritics(text)
+        return _phonikud(diacritizer_model).add_diacritics(text)
     if family == "ar":
-        return _tashkeel(model or diacritizer_model).diacritize(text)
+        return _tashkeel(diacritizer_model).diacritize(text)
     if family == "stress":
-        return _stress(text, lang, model)
+        return _stress(text, lang, diacritizer_model)
     if family == "pt":
         return _sense_diacritics_pt(text)
     return text
@@ -359,7 +366,8 @@ def register(graph) -> None:
     """
     graph.register(
         Edge("text", DIACRITIZED,
-             lambda text, lang="und", model=None, **c: diacritize(text, lang, model, **c),
+             lambda text, lang="und", diacritizer_model=None, **c:
+                 diacritize(text, lang, diacritizer_model=diacritizer_model, **c),
              lossless=False))
     graph.register(
         Edge(DIACRITIZED, "text",

{scriptconv-0.0.4a7 → scriptconv-0.0.4a8}/scriptconv/phonemizers/registry.py

@@ -231,9 +231,13 @@ def register(graph) -> None:
     from scriptconv.graph import Edge
 
     def _text_to_ipa(text: str, lang: str = "und",
-                     override: Optional[Phonemizer] = None, **_):
-        return phonemizer_for_lang(lang, Alphabet.IPA,
-                                   override).phonemize_string(text, lang)
+                     override: Optional[Phonemizer] = None,
+                     phonemizer_model: Optional[str] = None, **_):
+        # phonemizer_model is the model knob for this (phonemizer) edge —
+        # forwarded to model-backed phonemizers (ByT5/Charsiu/…); parallel to
+        # diacritizer_model on the text -> text-diacritized edge.
+        return phonemizer_for_lang(lang, Alphabet.IPA, override,
+                                   phonemizer_model).phonemize_string(text, lang)
 
     graph.register(Edge("text", "ipa", _text_to_ipa, lossless=False))
     graph.register(Edge("text-diacritized", "ipa", _text_to_ipa, lossless=False))

{scriptconv-0.0.4a7 → scriptconv-0.0.4a8}/scriptconv/version.py

@@ -2,7 +2,7 @@
 VERSION_MAJOR = 0
 VERSION_MINOR = 0
 VERSION_BUILD = 4
-VERSION_ALPHA = 7
+VERSION_ALPHA = 8
 # END_VERSION_BLOCK
 
 VERSION_STR = f"{VERSION_MAJOR}.{VERSION_MINOR}.{VERSION_BUILD}"

{scriptconv-0.0.4a7 → scriptconv-0.0.4a8/scriptconv.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: scriptconv
-Version: 0.0.4a7
+Version: 0.0.4a8
 Summary: Zero-dependency script & phoneme-notation core — ISO-15924 detection & metadata, IPA↔ARPABET/X-SAMPA/Lexique/Kirshenbaum/Cotovía/RFE, Buckwalter↔Arabic, Hangul→jamo, kana
 License: Apache-2.0
 Project-URL: Homepage, https://github.com/TigreGotico/scriptconv

{scriptconv-0.0.4a7 → scriptconv-0.0.4a8}/tests/test_diacritics_graph.py

@@ -101,3 +101,49 @@ class TestDiacriticsGraphExtension(unittest.TestCase):
 
 if __name__ == "__main__":
     unittest.main()
+
+
+class TestEdgeModelKeys(unittest.TestCase):
+    """The two model-bearing edges each read their own context key: the
+    diacritizer edge (text -> text-diacritized) reads ``diacritizer_model``,
+    the phonemizer edge (text -> ipa) reads ``phonemizer_model``. Neither picks
+    up the other's key."""
+
+    def setUp(self):
+        self.graph = DEFAULT_GRAPH.extend(diacritics.register).extend(phonemizers.register)
+
+    def test_diacritizer_edge_reads_diacritizer_model(self):
+        seen = {}
+        backend = types.SimpleNamespace(diacritize=lambda t: t)
+
+        def _spy(m=None):
+            seen["m"] = m
+            return backend
+
+        with mock.patch.object(diacritics, "_tashkeel", _spy):
+            self.graph.convert("نص", "text", "text-diacritized", lang="ar",
+                               diacritizer_model="RAWI-X",
+                               phonemizer_model="IGNORED")
+        self.assertEqual(seen["m"], "RAWI-X")
+
+    def test_phonemizer_edge_reads_phonemizer_model(self):
+        from scriptconv.phonemizers import registry
+        cap = {}
+
+        class _P:
+            def phonemize_string(self, t, l):
+                return "IPA"
+
+        def _spy(lang, alphabet, override=None, model=None, **k):
+            cap["model"] = model
+            return _P()
+
+        with mock.patch.object(registry, "phonemizer_for_lang", _spy):
+            self.graph.convert("hello", "text", "ipa", lang="en",
+                               phonemizer_model="BYT5-CKPT",
+                               diacritizer_model="IGNORED")
+        self.assertEqual(cap["model"], "BYT5-CKPT")
+
+
+if __name__ == "__main__":
+    unittest.main()