Refactor sentence processing in `text_generator.py`: centralize sentence end mark sets, streamline `_should_end_sentence` logic, and eliminate redundant parameter passing for improved clarity and maintainability.

src/VoiceDialogue/services/text/text_generator.py

@@ -35,6 +35,10 @@ class LLMResponseGenerator(BaseThread):
         self.user_question_queue = user_question_queue
         self.generated_answer_queue = generated_answer_queue
 
+        self.english_sentence_end_marks = {'!', '?', '.', ',', ':', ';'}
+        self.chinese_sentence_end_marks = {'，', '。', '！', '？', '：', '；', '、'}
+        self.sentence_end_marks = self.english_sentence_end_marks | self.chinese_sentence_end_marks
+
     def _get_prompt_by_language(self, language: str) -> str:
         """根据语言获取对应的 prompt"""
         if language == "zh":
@@ -64,19 +68,24 @@ class LLMResponseGenerator(BaseThread):
         messages = memory.load_memory_variables({})[key]
         return InMemoryChatMessageHistory(messages=messages)
 
-    def _should_end_sentence(self, sentence: str, sentence_end_mark: str,
-                             sentence_end_marks: set, is_first_sentence: bool) -> bool:
+    def _should_end_sentence(self, sentence: str, sentence_end_mark: str, is_first_sentence: bool) -> bool:
         """判断是否应该结束当前句子"""
-        if not sentence or sentence_end_mark not in sentence_end_marks:
+        if not sentence or sentence_end_mark not in self.sentence_end_marks:
             return False
 
+        is_chinese_sentence = False
+        if sentence_end_mark in self.chinese_sentence_end_marks:
+            is_chinese_sentence = True
+
         # 第一个句子的特殊处理逻辑
         if is_first_sentence:
-            chinese_sentence_end_marks = {'，', '。', '！', '？', '：', '；', '、'}
-            return (len(sentence) > 2 and sentence_end_mark in chinese_sentence_end_marks)
+            if is_chinese_sentence:
+                return (len(sentence) > 2 and sentence_end_mark in self.chinese_sentence_end_marks)
+            else:
+                return (len(sentence.split()) > 1 and sentence_end_mark in self.english_sentence_end_marks)
 
         # 普通句子的判断逻辑
-        if sentence_end_mark in {'，', '。', '！', '？', '：', '；', '、'}:
+        if is_chinese_sentence:
             sentence_words = len(sentence)
         else:
             sentence_words = len(sentence.split())
@@ -105,9 +114,6 @@ class LLMResponseGenerator(BaseThread):
 
     def _process_voice_task(self, voice_task: VoiceTask) -> None:
         """处理单个语音任务"""
-        english_sentence_end_marks = {'!', '?', '.', ',', ':', ';'}
-        chinese_sentence_end_marks = {'，', '。', '！', '？', '：', '；', '、'}
-        sentence_end_marks = english_sentence_end_marks | chinese_sentence_end_marks
 
         chunks = []
         answer_index = 0
@@ -124,9 +130,9 @@ class LLMResponseGenerator(BaseThread):
 
         try:
             for chunk in pipeline.stream(input={'input': user_question}, config=config):
-                chunk_content = f'{chunk.content.strip()}'
-                if not chunk_content:
+                if not chunk.content.strip():
                     continue
+                chunk_content = f'{chunk.content}'
 
                 sentence_end_mark, remain_content = self._process_chunk_content(chunk_content)
                 chunks.append(sentence_end_mark)
@@ -136,7 +142,7 @@ class LLMResponseGenerator(BaseThread):
                     continue
 
                 # 检查是否应该结束当前句子
-                if self._should_end_sentence(sentence, sentence_end_mark, sentence_end_marks, is_first_sentence):
+                if self._should_end_sentence(sentence, sentence_end_mark, is_first_sentence):
                     self._send_sentence_to_queue(voice_task, sentence, answer_index)
                     chunks = self._reset_chunks(remain_content)
                     answer_index += 1
@@ -146,19 +152,18 @@ class LLMResponseGenerator(BaseThread):
                         chunks.append(remain_content)
 
             # 处理最后剩余的 chunks
-            self._handle_remaining_chunks(voice_task, chunks, answer_index, sentence_end_marks)
+            self._handle_remaining_chunks(voice_task, chunks, answer_index)
 
         except Exception as e:
             print(f'处理语音任务时发生错误: {e}')
 
-    def _handle_remaining_chunks(self, voice_task: VoiceTask, chunks: list,
-                                 answer_index: int, sentence_end_marks: set) -> None:
+    def _handle_remaining_chunks(self, voice_task: VoiceTask, chunks: list, answer_index: int) -> None:
         """处理剩余的 chunks"""
         if not chunks:
             return
 
         sentence = preprocess_sentence_text(chunks)
-        if not sentence or sentence.strip() in sentence_end_marks:
+        if not sentence or sentence.strip() in self.sentence_end_marks:
             return
 
         self._send_sentence_to_queue(voice_task, sentence, answer_index)