scriptconv 0.0.4a10__tar.gz → 0.0.4a11__tar.gz

{scriptconv-0.0.4a10/scriptconv.egg-info → scriptconv-0.0.4a11}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: scriptconv
-Version: 0.0.4a10
+Version: 0.0.4a11
 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-0.0.4a10 → scriptconv-0.0.4a11}/scriptconv/phonemizers/ru.py

@@ -5,7 +5,7 @@ Currently one backend: the Vosk-TTS front-end, wrapping the vendored
 Russian voices can be driven from text without the ``vosk-tts`` package.
 """
 import re
-from typing import List, Optional
+from typing import Iterator, List, Optional
 
 from quebra_frases import sentence_tokenize
 
@@ -86,25 +86,28 @@ class VoskPhonemizer(BasePhonemizer):
                 tokens.extend(convert(word).split())
         return tokens
 
-    def phonemize(self, text: str, lang: str) -> PhonemizedChunks:
-        """Sentence-level lists of Vosk phoneme tokens.
+    def phonemize_lazy(self, text: str, lang: str) -> Iterator[List[str]]:
+        """Sentence-level lists of Vosk phoneme tokens, one sentence at a time.
 
         Punctuation is preserved (it drives pausing); each sentence becomes one
         synthesis chunk.  Multi-character tokens (``sch``, ``bj``, ``a1``) stay
-        whole, so :meth:`BasePhonemizer.phonemize`'s per-character split is
-        deliberately bypassed.
+        whole, so :meth:`BasePhonemizer.phonemize_lazy`'s per-character split of
+        the phoneme *string* is deliberately bypassed — splitting ``s h`` into
+        characters would later fold back into ``sh`` (ш), a different phoneme.
         """
         self.get_lang(lang)
         if not text:
-            return []
+            return
         if self.normalizer is not None:
             text = self.normalizer(text, lang)
-        results: PhonemizedChunks = []
         for sentence in sentence_tokenize(text):
             tokens = self._g2p_tokens(sentence)
             if tokens:
-                results.append(tokens)
-        return results
+                yield tokens
+
+    def phonemize(self, text: str, lang: str) -> PhonemizedChunks:
+        """Sentence-level lists of Vosk phoneme tokens."""
+        return list(self.phonemize_lazy(text, lang))
 
     def phonemize_to_list(self, text: str, lang: str) -> List[str]:
         self.get_lang(lang)

{scriptconv-0.0.4a10 → scriptconv-0.0.4a11}/scriptconv/version.py

@@ -2,7 +2,7 @@
 VERSION_MAJOR = 0
 VERSION_MINOR = 0
 VERSION_BUILD = 4
-VERSION_ALPHA = 10
+VERSION_ALPHA = 11
 # END_VERSION_BLOCK
 
 VERSION_STR = f"{VERSION_MAJOR}.{VERSION_MINOR}.{VERSION_BUILD}"

{scriptconv-0.0.4a10 → scriptconv-0.0.4a11/scriptconv.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: scriptconv
-Version: 0.0.4a10
+Version: 0.0.4a11
 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-0.0.4a10 → scriptconv-0.0.4a11}/tests/test_phonemizers_ru.py

@@ -209,6 +209,21 @@ class TestVoskRegistration(unittest.TestCase):
         p = phonemizer_for_lang("ru", alphabet=Alphabet.IPA)
         self.assertNotIsInstance(p, VoskPhonemizer)
 
+    def test_phonemize_lazy_keeps_tokens_whole(self):
+        # the lazy path is what consumers stream through: it must yield the same
+        # whole phoneme tokens as phonemize(), never a per-character split
+        # ("s", "h" would otherwise fold back into "sh" — a different phoneme)
+        p = VoskPhonemizer()
+        text = "Сходить в кино. Счастье рядом!"
+        self.assertEqual(list(p.phonemize_lazy(text, "ru")), p.phonemize(text, "ru"))
+        flat = [t for chunk in p.phonemize_lazy(text, "ru") for t in chunk]
+        self.assertIn("s", flat)
+        self.assertIn("h", flat)
+        self.assertTrue(all(len(t) <= 3 for t in flat))
+        # every emitted token is a real phoneme/pause, not a bare character of one
+        self.assertNotIn("0", flat)
+        self.assertNotIn("1", flat)
+
     def test_module_level_phonemize_facade(self):
         self.assertEqual(phonemize("Привет", "ru", alphabet=Alphabet.VOSK),
                          "p rj i0 vj e0 t")