scriptconv 0.0.3a14__tar.gz → 0.0.3a16__tar.gz

{scriptconv-0.0.3a14 → scriptconv-0.0.3a16}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: scriptconv
-Version: 0.0.3a14
+Version: 0.0.3a16
 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-0.0.3a14 → scriptconv-0.0.3a16}/scriptconv/phonemizers/base.py

@@ -45,8 +45,9 @@ class BasePhonemizer(metaclass=abc.ABCMeta):
         # optional (text, lang) -> str hook run before chunking; see module
         # docstring — scriptconv performs no normalization of its own
         self.normalizer = normalizer
-        # local path to a phonikud ONNX model (Hebrew diacritization);
-        # scriptconv never downloads — the consumer resolves the file
+        # local path to a phonikud ONNX model (Hebrew diacritization), or a
+        # zero-arg callable resolving one lazily; scriptconv never downloads
+        # — the consumer resolves the file
         self.phonikud_model = phonikud_model
 
         # diacritizer model name, for languages that need one. Arabic uses
@@ -59,7 +60,9 @@ class BasePhonemizer(metaclass=abc.ABCMeta):
     @property
     def phonikud(self):
         if self._phonikud is None:
-            if not self.phonikud_model:
+            model = self.phonikud_model() if callable(self.phonikud_model) \
+                else self.phonikud_model
+            if not model:
                 raise ValueError(
                     "Hebrew diacritization needs a local phonikud ONNX model: "
                     "pass phonikud_model=<path> (scriptconv never downloads "
@@ -70,7 +73,7 @@ class BasePhonemizer(metaclass=abc.ABCMeta):
                 raise ImportError(
                     "Hebrew diacritization needs phonikud-onnx — install "
                     "with `pip install scriptconv[he]`") from None
-            self._phonikud = Phonikud(self.phonikud_model)
+            self._phonikud = Phonikud(model)
         return self._phonikud
 
     def tashkeel(self, model: Optional[str] = None):

{scriptconv-0.0.3a14 → scriptconv-0.0.3a16}/scriptconv/phonemizers/registry.py

@@ -133,30 +133,61 @@ _EMITS: Dict[Phonemizer, Tuple[Alphabet, ...]] = {
 }
 
 LANG_DEFAULTS: Dict[str, Tuple[Phonemizer, ...]] = {
-    # Arabic IPA always goes through arbtok (hard org rule)
+    # Arabic IPA always goes through arbtok (hard org rule — never falls
+    # through to orthography2ipa or espeak)
     "ar": (_P.ARBTOK,),
     "eu": (_P.EUSKAPHONE,),
     "mwl": (_P.MIRANDESE,),
     "pt": (_P.TUGAPHONE,),
     "he": (_P.PHONIKUD,),
-    "gl": (_P.COTOVIA, _P.ESPEAK),
+    "gl": (_P.COTOVIA, _P.ORTHOGRAPHY2IPA, _P.ESPEAK),
 }
 
+# Languages whose explicit entry is exhaustive: no generic fallback beyond it.
+_NO_GENERIC_FALLBACK = {"ar"}
+
+
+def _o2i_supports(lang: str) -> bool:
+    """True when orthography2ipa is installed and has a spec for *lang*."""
+    try:
+        from scriptconv.phonemizers.o2ipa import Orthography2IPAPhonemizer
+        Orthography2IPAPhonemizer.get_lang(lang)
+        return True
+    except (ImportError, ValueError):
+        return False
+
 
 def phonemizer_for_lang(lang: str, alphabet: Alphabet = Alphabet.IPA,
                         override: Optional[Phonemizer] = None,
                         model: Optional[str] = None, **kwargs):
     """Construct the default phonemizer for *lang* (or the override).
 
-    Candidates from :data:`LANG_DEFAULTS` are filtered by the alphabet they
-    can emit; espeak is the universal fallback.
+    Resolution order — in-house engines first:
+
+    1. ``override=`` when given;
+    2. the language's explicit :data:`LANG_DEFAULTS` chain, filtered by the
+       alphabet each candidate can emit;
+    3. :data:`~Phonemizer.ORTHOGRAPHY2IPA` when orthography2ipa supports the
+       language (the usual default);
+    4. espeak as the last resort.
+
+    Languages in the no-fallback set (Arabic) never leave their explicit
+    chain — a missing arbtok raises rather than silently degrading.
     """
     if override is not None:
         return get_phonemizer(override, alphabet, model, **kwargs)
     key = lang.replace("_", "-").split("-")[0].lower()
-    for candidate in LANG_DEFAULTS.get(key, ()):
+    explicit = LANG_DEFAULTS.get(key, ())
+    for candidate in explicit:
         if alphabet in _EMITS.get(candidate, (Alphabet.IPA,)):
             return get_phonemizer(candidate, alphabet, model, **kwargs)
+    if key in _NO_GENERIC_FALLBACK:
+        raise ValueError(
+            f"no eligible phonemizer for {lang!r} with alphabet "
+            f"{alphabet.value!r}: the explicit chain "
+            f"{[c.value for c in explicit]} is exhaustive for this language")
+    if alphabet == Alphabet.IPA and _o2i_supports(lang):
+        return get_phonemizer(_P.ORTHOGRAPHY2IPA, alphabet, model, **kwargs)
     return get_phonemizer(_P.ESPEAK, alphabet, model, **kwargs)
 
 

{scriptconv-0.0.3a14 → scriptconv-0.0.3a16}/scriptconv/version.py

@@ -2,7 +2,7 @@
 VERSION_MAJOR = 0
 VERSION_MINOR = 0
 VERSION_BUILD = 3
-VERSION_ALPHA = 14
+VERSION_ALPHA = 16
 # END_VERSION_BLOCK
 
 VERSION_STR = f"{VERSION_MAJOR}.{VERSION_MINOR}.{VERSION_BUILD}"

{scriptconv-0.0.3a14 → scriptconv-0.0.3a16}/scriptconv.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: scriptconv
-Version: 0.0.3a14
+Version: 0.0.3a16
 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-0.0.3a14 → scriptconv-0.0.3a16}/tests/test_phonemizers_base.py

@@ -96,12 +96,34 @@ class TestLangDefaults(unittest.TestCase):
         self.assertEqual(LANG_DEFAULTS["ar"], (Phonemizer.ARBTOK,))
 
     def test_cotovia_only_for_cotovia_alphabet(self):
+        # requesting IPA must never yield Cotovía (it emits its own notation);
+        # the in-house chain falls through to orthography2ipa (or espeak)
         gl_ipa = phonemizer_for_lang("gl", alphabet=Alphabet.IPA)
-        self.assertEqual(type(gl_ipa).__name__, "EspeakPhonemizer")
-
-    def test_fallback_is_espeak(self):
-        de = phonemizer_for_lang("de-DE")
-        self.assertEqual(type(de).__name__, "EspeakPhonemizer")
+        self.assertIn(type(gl_ipa).__name__,
+                      ("Orthography2IPAPhonemizer", "EspeakPhonemizer"))
+        self.assertNotEqual(type(gl_ipa).__name__, "CotoviaPhonemizer")
+        gl_ctv = phonemizer_for_lang("gl", alphabet=Alphabet.COTOVIA)
+        self.assertEqual(type(gl_ctv).__name__, "CotoviaPhonemizer")
+
+    def test_fallback_prefers_o2i_then_espeak(self):
+        # in-house first: orthography2ipa is the usual default when it has a
+        # spec for the language; espeak only as last resort
+        from unittest import mock
+        import scriptconv.phonemizers.registry as reg
+        with mock.patch.object(reg, "_o2i_supports", return_value=True):
+            with mock.patch.object(reg, "get_phonemizer") as gp:
+                reg.phonemizer_for_lang("de-DE")
+                self.assertEqual(gp.call_args[0][0], Phonemizer.ORTHOGRAPHY2IPA)
+        with mock.patch.object(reg, "_o2i_supports", return_value=False):
+            de = reg.phonemizer_for_lang("de-DE")
+            self.assertEqual(type(de).__name__, "EspeakPhonemizer")
+
+    def test_arabic_never_falls_through(self):
+        # hard org rule: Arabic IPA only via arbtok — with arbtok ineligible
+        # (wrong alphabet), the resolver raises instead of degrading to
+        # o2i/espeak
+        with self.assertRaises((ValueError, ImportError)):
+            phonemizer_for_lang("ar", alphabet=Alphabet.ARPA)
 
     def test_override_wins(self):
         g = phonemizer_for_lang("de", override=Phonemizer.GRAPHEMES)
@@ -159,3 +181,18 @@ class TestFacadeAndGraph(unittest.TestCase):
         from scriptconv import phonemizers
         g = DEFAULT_GRAPH.extend(phonemizers.register)
         self.assertTrue(g.can_convert("text", "arpa"))
+
+
+class TestPhonikudModelResolver(unittest.TestCase):
+    def test_callable_resolver_invoked_lazily(self):
+        calls = []
+
+        def resolver():
+            calls.append(1)
+            return ""  # resolves to nothing -> still the explicit ValueError
+
+        g = GraphemePhonemizer(phonikud_model=resolver)
+        self.assertEqual(calls, [])  # not resolved at construction
+        with self.assertRaises(ValueError):
+            g.add_diacritics("שלום", "he")
+        self.assertEqual(calls, [1])