From a3c4e040c80136413aa9397a02fd5f2a45f10e73 Mon Sep 17 00:00:00 2001
From: KamioRinn <63162909+KamioRinn@users.noreply.github.com>
Date: Fri, 12 Apr 2024 11:18:34 +0800
Subject: [PATCH] Normalize chinese arithmetic operations (#947)

---
 GPT_SoVITS/text/zh_normalization/num.py       | 33 +++++++++++++++++--
 .../zh_normalization/text_normlization.py     |  9 ++++-
 2 files changed, 39 insertions(+), 3 deletions(-)

diff --git a/GPT_SoVITS/text/zh_normalization/num.py b/GPT_SoVITS/text/zh_normalization/num.py
index 8ef7f48..d38d5a6 100644
--- a/GPT_SoVITS/text/zh_normalization/num.py
+++ b/GPT_SoVITS/text/zh_normalization/num.py
@@ -106,6 +106,29 @@ def replace_default_num(match):
     return verbalize_digit(number, alt_one=True)
 
 
+# 加减乘除
+RE_ASMD = re.compile(
+    r'((-?)((\d+)(\.\d+)?)|(\.(\d+)))([\+\-\×÷=])((-?)((\d+)(\.\d+)?)|(\.(\d+)))')
+asmd_map = {
+    '+': '加',
+    '-': '减',
+    '×': '乘',
+    '÷': '除',
+    '=': '等于'
+}
+
+
+def replace_asmd(match) -> str:
+    """
+    Args:
+        match (re.Match)
+    Returns:
+        str
+    """
+    result = match.group(1) + asmd_map[match.group(8)] + match.group(9)
+    return result
+
+
 # 数字表达式
 # 纯小数
 RE_DECIMAL_NUM = re.compile(r'(-?)((\d+)(\.\d+))' r'|(\.(\d+))')
@@ -155,7 +178,13 @@ def replace_number(match) -> str:
 # match.group(1) and match.group(8) are copy from RE_NUMBER
 
 RE_RANGE = re.compile(
-    r'((-?)((\d+)(\.\d+)?)|(\.(\d+)))[-~]((-?)((\d+)(\.\d+)?)|(\.(\d+)))')
+    r"""
+    (?<![\d\+\-\×÷=])      # 使用反向前瞻以确保数字范围之前没有其他数字和操作符
+    ((-?)((\d+)(\.\d+)?))  # 匹配范围起始的负数或正数（整数或小数）
+    [-~]                   # 匹配范围分隔符
+    ((-?)((\d+)(\.\d+)?))  # 匹配范围结束的负数或正数（整数或小数）
+    (?![\d\+\-\×÷=])       # 使用正向前瞻以确保数字范围之后没有其他数字和操作符
+    """, re.VERBOSE)
 
 
 def replace_range(match) -> str:
@@ -165,7 +194,7 @@ def replace_range(match) -> str:
     Returns:
         str
     """
-    first, second = match.group(1), match.group(8)
+    first, second = match.group(1), match.group(6)
     first = RE_NUMBER.sub(replace_number, first)
     second = RE_NUMBER.sub(replace_number, second)
     result = f"{first}到{second}"
diff --git a/GPT_SoVITS/text/zh_normalization/text_normlization.py b/GPT_SoVITS/text/zh_normalization/text_normlization.py
index b4c1494..e852fe9 100644
--- a/GPT_SoVITS/text/zh_normalization/text_normlization.py
+++ b/GPT_SoVITS/text/zh_normalization/text_normlization.py
@@ -34,6 +34,7 @@ from .num import RE_PERCENTAGE
 from .num import RE_POSITIVE_QUANTIFIERS
 from .num import RE_RANGE
 from .num import RE_TO_RANGE
+from .num import RE_ASMD
 from .num import replace_default_num
 from .num import replace_frac
 from .num import replace_negative_num
@@ -42,6 +43,7 @@ from .num import replace_percentage
 from .num import replace_positive_quantifier
 from .num import replace_range
 from .num import replace_to_range
+from .num import replace_asmd
 from .phonecode import RE_MOBILE_PHONE
 from .phonecode import RE_NATIONAL_UNIFORM_NUMBER
 from .phonecode import RE_TELEPHONE
@@ -67,7 +69,7 @@ class TextNormalizer():
         if lang == "zh":
             text = text.replace(" ", "")
             # 过滤掉特殊字符
-            text = re.sub(r'[——《》【】<=>{}()（）#&@“”^_|\\]', '', text)
+            text = re.sub(r'[——《》【】<>{}()（）#&@“”^_|\\]', '', text)
         text = self.SENTENCE_SPLITOR.sub(r'\1\n', text)
         text = text.strip()
         sentences = [sentence.strip() for sentence in re.split(r'\n+', text)]
@@ -142,6 +144,11 @@ class TextNormalizer():
         sentence = RE_NATIONAL_UNIFORM_NUMBER.sub(replace_phone, sentence)
 
         sentence = RE_RANGE.sub(replace_range, sentence)
+
+        # 处理加减乘除
+        while RE_ASMD.search(sentence):
+            sentence = RE_ASMD.sub(replace_asmd, sentence)
+
         sentence = RE_INTEGER.sub(replace_negative_num, sentence)
         sentence = RE_DECIMAL_NUM.sub(replace_number, sentence)
         sentence = RE_POSITIVE_QUANTIFIERS.sub(replace_positive_quantifier,