Fix generating Chinese lexicon for Kokoro TTS 1.0 (#1888)

@@ -2,10 +2,21 @@
 # Copyright    2025  Xiaomi Corp.        (authors: Fangjun Kuang)
 # Copyright    2025  Xiaomi Corp.        (authors: Fangjun Kuang)
 
 
 import json
 import json
-from pypinyin import phrases_dict, pinyin_dict
-from misaki import zh
 from typing import List, Tuple
 from typing import List, Tuple
 
 
+from misaki import zh
+from pypinyin import load_phrases_dict, phrases_dict, pinyin_dict
+
+user_dict = {
+    "还田": [["huan2"], ["tian2"]],
+    "行长": [["hang2"], ["zhang3"]],
+    "银行行长": [["yin2"], ["hang2"], ["hang2"], ["zhang3"]],
+}
+
+load_phrases_dict(user_dict)
+
+phrases_dict.phrases_dict.update(**user_dict)
+
 
 
 def generate_english_lexicon(kind: str):
 def generate_english_lexicon(kind: str):
     assert kind in ("us", "gb"), kind
     assert kind in ("us", "gb"), kind
@@ -59,11 +70,13 @@ def generate_chinese_lexicon():
         if not (0x4E00 <= key <= 0x9FFF):
         if not (0x4E00 <= key <= 0x9FFF):
             continue
             continue
         w = chr(key)
         w = chr(key)
-        tokens: str = g2p(w)
+        tokens: str = g2p.word2ipa(w)
+        tokens = tokens.replace(chr(815), "")
         lexicon.append((w, tokens))
         lexicon.append((w, tokens))
 
 
     for key in phrases:
     for key in phrases:
-        tokens: str = g2p(key)
+        tokens: str = g2p.word2ipa(key)
+        tokens = tokens.replace(chr(815), "")
         lexicon.append((key, tokens))
         lexicon.append((key, tokens))
     return lexicon
     return lexicon
 
 

@@ -114,11 +114,6 @@ if [ ! -f ./lexicon-zh.txt ]; then
   ./generate_lexicon.py
   ./generate_lexicon.py
 fi
 fi
 
 
-grep '还钱' ./lexicon-zh.txt
-sed -i.bak 's/还钱 x a i/还钱 x w a/' ./lexicon-zh.txt
-rm -v ./lexicon-zh.txt.bak
-grep '还钱' ./lexicon-zh.txt
-
 if [ ! -f ./voices.bin ]; then
 if [ ! -f ./voices.bin ]; then
   ./generate_voices_bin.py
   ./generate_voices_bin.py
 fi
 fi