Merge 8c0cb0d691554d8311d8904972ae3efa8bfd1cc4 into b2cff0cd0abd0ac134a16ae7a9695f88e8826104

* fix: 固定唑的读音

* add: 增加md5检测以更新缓存

GPT_SoVITS/text/cleaner.py

@@ -10,14 +10,14 @@ import os
 from text import symbols as symbols_v1
 from text import symbols2 as symbols_v2
 
+
 special = [
     # ("%", "zh", "SP"),
-    ("￥", "zh", "SP2"),
+    # ("￥", "zh", "SP2"), #加了货币计数所以人民币符不是SP2了
     ("^", "zh", "SP3"),
     # ('@', 'zh', "SP4")#不搞鬼畜了，和第二版保持一致吧
 ]
 
-
 def clean_text(text, language, version=None):
     if version is None:
         version = os.environ.get("version", "v2")
@@ -31,6 +31,14 @@ def clean_text(text, language, version=None):
     if language not in language_module_map:
         language = "en"
         text = " "
+    if language in ("zh"): #处理货币似乎最佳方案是这里截胡，不然可能被吞...
+        from text.zh_normalization.num import (
+            RE_CNY_PREFIX, RE_CNY_SUFFIX, replace_cny_prefix, replace_cny_suffix,
+            RE_USD_SYMBOL, RE_USD_SUFFIX, replace_usd_symbol, replace_usd_suffix,)
+        text = RE_CNY_PREFIX.sub(replace_cny_prefix, text)
+        text = RE_CNY_SUFFIX.sub(replace_cny_suffix, text)
+        text = RE_USD_SYMBOL.sub(replace_usd_symbol, text)
+        text = RE_USD_SUFFIX.sub(replace_usd_suffix, text)
     for special_s, special_l, target_symbol in special:
         if special_s in text and language == special_l:
             return clean_special(text, language, special_s, target_symbol, version)

GPT_SoVITS/text/g2pw/g2pw.py

@@ -1,5 +1,6 @@
 # This code is modified from https://github.com/mozillazg/pypinyin-g2pW
 
+import hashlib
 import pickle
 import os
 
@@ -14,6 +15,16 @@ current_file_path = os.path.dirname(__file__)
 CACHE_PATH = os.path.join(current_file_path, "polyphonic.pickle")
 PP_DICT_PATH = os.path.join(current_file_path, "polyphonic.rep")
 PP_FIX_DICT_PATH = os.path.join(current_file_path, "polyphonic-fix.rep")
+MD5_PATH = os.path.join(current_file_path, "polyphonic.md5")
+
+def get_file_md5(file_path):
+    if not os.path.exists(file_path):
+        return ""
+    hasher = hashlib.md5()
+    with open(file_path, "rb") as f:
+        for chunk in iter(lambda: f.read(4096), b""):
+            hasher.update(chunk)
+    return hasher.hexdigest()
 
 
 class G2PWPinyin(Pinyin):
@@ -115,13 +126,22 @@ def cache_dict(polyphonic_dict, file_path):
 
 
 def get_dict():
-    if os.path.exists(CACHE_PATH):
+    new_md5 = get_file_md5(PP_DICT_PATH) + get_file_md5(PP_FIX_DICT_PATH)
+    old_md5 = ""
+    if os.path.exists(MD5_PATH):
+        with open(MD5_PATH, "r", encoding="utf-8") as f:
+            old_md5 = f.read().strip()
+    need_rebuild = (not os.path.exists(CACHE_PATH)) or (new_md5 != old_md5)
+
+    if not need_rebuild:
         with open(CACHE_PATH, "rb") as pickle_file:
             polyphonic_dict = pickle.load(pickle_file)
     else:
+        print("Rebuilding Polyphonic Dictionary: " + f"{old_md5} -> {new_md5}")
         polyphonic_dict = read_dict()
         cache_dict(polyphonic_dict, CACHE_PATH)
-
+        with open(MD5_PATH, "w", encoding="utf-8") as f:
+            f.write(new_md5)
     return polyphonic_dict
 
 

GPT_SoVITS/text/g2pw/polyphonic-fix.rep

@@ -45023,4 +45023,5 @@
 鼎铛玉石: ['ding3', 'cheng1', 'yu4', 'shi2']
 齿豁头童: ['chi3', 'huo1', 'tou2', 'tong2']
 牦牛: ['mao2', 'niu2']
-牦: ['mao2']
+牦: ['mao2']
+唑: ['zuo4']

GPT_SoVITS/text/g2pw/polyphonic.md5

@@ -0,0 +1 @@
+13b2211c317c75794123ffdf7c2aea021c75b0c606ad61d7c8b05bb00b64fa21

GPT_SoVITS/text/zh_normalization/num.py

@@ -337,3 +337,116 @@ def num2str(value_string: str) -> str:
         result = result if result else "零"
         result += "点" + verbalize_digit(decimal)
     return result
+
+RE_CNY_PREFIX = re.compile(r"(?:¥|￥)\s*(-?\d[\d,]*(?:\.\d+)?)")
+RE_CNY_SUFFIX = re.compile(r"(-?\d[\d,]*(?:\.\d+)?)(?:\s*(?:人民币|元|CNY|cny|¥|￥))")
+
+def _strip_commas(s: str) -> str:
+    return s.replace(",", "")
+
+def _split_amount(amount: str):
+    neg = amount.startswith("-")
+    if neg:
+        amount = amount[1:]
+    amount = _strip_commas(amount) or "0"
+
+    if "." in amount:
+        integer, frac = amount.split(".", 1)
+        had_frac = True
+    else:
+        integer, frac, had_frac = amount, "", False
+
+    integer = integer or "0"
+    frac = (frac + "00")[:2]
+    return neg, integer, frac, had_frac
+
+#人民币和美元的处理都在cleaner那边，防吞
+def replace_cny_amount(amount: str, num2str) -> str:
+    neg, integer, frac, had_frac = _split_amount(amount)
+
+    integer_cn = num2str(integer) if integer != "0" else "零"
+
+    jiao, fen = frac[0], frac[1]
+    parts = []
+
+    if integer != "0":
+        parts.append(integer_cn + "元")
+    else:
+        parts.append("零元")
+
+    if jiao != "0" or fen != "0":
+        if jiao != "0":
+            parts.append(num2str(jiao) + "角")
+        if fen != "0":
+            parts.append(num2str(fen) + "分")
+    elif had_frac:
+        parts.append("整")
+
+    res = "".join(parts)
+    if neg and res and res[0] != "负":
+        res = "负" + res
+    return res
+
+def replace_cny_prefix(m, num2str=num2str):
+    return replace_cny_amount(m.group(1), num2str)
+
+def replace_cny_suffix(m, num2str=num2str):
+    return replace_cny_amount(m.group(1), num2str)
+
+#我知道美元符也可能是加拿大元什么的，但是就当它美元吧whatever
+RE_USD_SYMBOL = re.compile(r"(?:\$|＄)\s*(-?\d[\d,]*(?:\.\d+)?)")
+RE_USD_SUFFIX = re.compile(r"(-?\d[\d,]*(?:\.\d+)?)(?:\s*(?:美元|USD|usd|\$|＄))")
+
+def _strip_commas(s: str) -> str:
+    return s.replace(",", "")
+
+def _split_amount(amount: str):
+    neg = amount.startswith("-")
+    if neg:
+        amount = amount[1:]
+    amount = _strip_commas(amount) or "0"
+
+    if "." in amount:
+        integer, frac = amount.split(".", 1)
+        had_frac = True
+    else:
+        integer, frac, had_frac = amount, "", False
+
+    integer = integer or "0"
+    # 只保留两位小数用来读美分
+    frac = (frac + "00")[:2]
+    return neg, integer, frac, had_frac
+
+def replace_usd_amount(amount: str, num2str) -> str:
+    neg, integer, frac, had_frac = _split_amount(amount)
+
+    integer_cn = num2str(integer) if integer != "0" else "零"
+
+    jiao, fen = frac[0], frac[1]
+    parts = []
+    if integer != "0":
+        parts.append(integer_cn + "美元")
+
+    if jiao != "0" or fen != "0":
+        cents = ""
+        if jiao != "0":
+            cents += num2str(jiao) + "十"
+        if fen != "0":
+            cents += num2str(fen)
+        cents = cents.replace("一十", "十")
+        parts.append(cents + "美分")
+    elif had_frac:
+        parts.append("整")
+    elif integer == "0":
+        parts = ["零美元"]
+
+    res = "".join(parts)
+    if neg and res and res[0] != "负":
+        res = "负" + res
+    return res
+
+def replace_usd_symbol(m, num2str=num2str):
+    return replace_usd_amount(m.group(1), num2str)
+
+def replace_usd_suffix(m, num2str=num2str):
+    return replace_usd_amount(m.group(1), num2str)