重构处理LLM生成chunk内容的方法，以提高文本生成的准确性。

src/voice_dialogue/services/text/generator.py

@@ -1,5 +1,6 @@
 import copy
 import time
+import unicodedata
 from queue import Queue, Empty
 
 from langchain.memory import ConversationBufferWindowMemory
@@ -99,12 +100,30 @@ class LLMResponseGenerator(BaseThread):
         """重置 chunks 列表"""
         return [remain_content] if remain_content else []
 
+    def _is_punctuation(self, char: str) -> bool:
+        """判断一个字符是否是标点符号"""
+        if not char or len(char) != 1:
+            return False
+        # 检查字符的 Unicode 类别是否为标点符号 (Punctuation)
+        return unicodedata.category(char).startswith('P')
+
     def _process_chunk_content(self, chunk_content: str) -> tuple:
-        """处理 chunk 内容，分离句子结束标记和剩余内容"""
-        if len(chunk_content) > 1:
-            return chunk_content[0], chunk_content[1:]
-        else:
-            return chunk_content, ''
+        """处理 chunk 内容，从右到左找标点符号并分割成三部分"""
+        if not chunk_content:
+            return '', '', ''
+
+        # 从右到左迭代，找到第一个标点符号
+        for i in range(len(chunk_content) - 1, -1, -1):
+            char = chunk_content[i]
+            if self._is_punctuation(char):
+                # 找到标点符号，分割成三部分
+                before_punct = chunk_content[:i]  # 标点之前的部分
+                punct = char  # 标点符号本身
+                after_punct = chunk_content[i + 1:]  # 标点之后的部分
+                return before_punct, punct, after_punct
+
+        # 如果没有找到标点符号，返回整个内容作为第一部分
+        return chunk_content, '', ''
 
     def _process_voice_task(self, voice_task: VoiceTask) -> None:
         """处理单个语音任务"""
@@ -140,8 +159,11 @@ class LLMResponseGenerator(BaseThread):
 
                 chunk_content = f'{chunk.content}'
 
-                sentence_end_mark, remain_content = self._process_chunk_content(chunk_content)
-                chunks.append(sentence_end_mark)
+                before_punct, sentence_end_mark, remain_content = self._process_chunk_content(chunk_content)
+                if before_punct:
+                    chunks.append(before_punct)
+                if sentence_end_mark:
+                    chunks.append(sentence_end_mark)
 
                 sentence = preprocess_sentence_text(chunks)
                 if not sentence: