Upload pronunciation.py with huggingface_hub

pronunciation.py

@@ -0,0 +1,87 @@
+"""
+pronunciation.py
+----------------
+Phonetic lexicon and pronunciation normalization for difficult Vedic, Sanskrit,
+and historical terminology when synthesizing speech via Kokoro-ONNX.
+
+Ensures proper pronunciation without confusing espeak-ng/phonemizer in English mode.
+"""
+
+import re
+from typing import Dict, Optional
+
+# Core Vedic & Sanskrit phonetic replacement lexicon for Kokoro (en-gb/en-us)
+VEDIC_PRONUNCIATION_MAP: Dict[str, str] = {
+    # Diacritic and IAST variations
+    r"\bṚṣis?\b": "Rishis",
+    r"\bRsis\b": "Rishis",
+    r"\bRsi\b": "Rishi",
+    r"\bṚta\b": "Ree-tuh",
+    r"\bRta\b": "Ree-tuh",
+    r"\bEkam Sat\b": "Ay-kum Sut",
+    r"\bTad Ekam\b": "Tud Ay-kum",
+    r"\bEkam\b": "Ay-kum",
+    r"\bNasadiya Sukta\b": "Nah-suh-dee-yuh Sook-tuh",
+    r"\bNāsadīya Sūkta\b": "Nah-suh-dee-yuh Sook-tuh",
+    r"\bVisvakarman\b": "Vish-vuh-kur-mun",
+    r"\bViśvakarman\b": "Vish-vuh-kur-mun",
+    r"\bPrajapati\b": "Pruh-jah-puh-tee",
+    r"\bPrajāpati\b": "Pruh-jah-puh-tee",
+    r"\bAgni\b": "Ugg-nee",
+    r"\bVaruna\b": "Vuh-roo-nuh",
+    r"\bVaruṇa\b": "Vuh-roo-nuh",
+    r"\bVayu\b": "Vah-yoo",
+    r"\bVāyu\b": "Vah-yoo",
+    r"\bVisnu\b": "Vishnu",
+    r"\bViṣṇu\b": "Vishnu",
+    r"\bBrihad-Aranyaka\b": "Bree-hud Ah-run-yuh-kuh",
+    r"\bBṛhadāraṇyaka\b": "Bree-hud Ah-run-yuh-kuh",
+    r"\bUpanishads\b": "Oo-pun-ee-shuds",
+    r"\bUpanishad\b": "Oo-pun-ee-shud",
+    r"\bUpaniṣads?\b": "Oo-pun-ee-shud",
+    r"\bYajnavalkya\b": "Yaj-nuh-val-kya",
+    r"\bYājñavalkya\b": "Yaj-nuh-val-kya",
+    r"\bmahiman\b": "muh-hee-mun",
+    r"\bBrahman\b": "Brah-mun",
+    r"\bBrahma\b": "Brah-muh",
+    r"\bAtman\b": "Aht-mun",
+    r"\bĀtman\b": "Aht-mun",
+    r"\bChandogya\b": "Chahn-dohg-yuh",
+    r"\bChāndogya\b": "Chahn-dohg-yuh",
+    r"\bKatha\b": "Kuh-tuh",
+    r"\bKaṭha\b": "Kuh-tuh",
+    r"\bMaitri\b": "My-tree",
+    r"\bMaitrī\b": "My-tree",
+    r"\bRig Veda\b": "Rig Vay-duh",
+    r"\bRigveda\b": "Rig Vay-duh",
+    r"\bṚgveda\b": "Rig Vay-duh",
+    r"\bVedas\b": "Vay-duhs",
+    r"\bVeda\b": "Vay-duh",
+    r"\bVedic\b": "Vay-dik",
+    r"\bSamsara\b": "Sum-sah-ruh",
+    r"\bSaṃsāra\b": "Sum-sah-ruh",
+    r"\bMoksha\b": "Mohk-shuh",
+    r"\bMokṣa\b": "Mohk-shuh",
+    r"\bMaya\b": "Mah-yah",
+    r"\bMāyā\b": "Mah-yah",
+    r"\bDharma\b": "Dhur-muh",
+    r"\bKarma\b": "Kur-muh",
+    r"\bSoma\b": "Soh-muh",
+}
+
+
+def apply_pronunciation_lexicon(
+    text: str, custom_map: Optional[Dict[str, str]] = None
+) -> str:
+    """
+    Replaces ancient and difficult terms with their phonetic equivalents for TTS.
+    """
+    active_map = VEDIC_PRONUNCIATION_MAP.copy()
+    if custom_map:
+        active_map.update(custom_map)
+
+    result = text
+    for pattern, replacement in active_map.items():
+        result = re.sub(pattern, replacement, result, flags=re.IGNORECASE)
+
+    return result