update:优化了websocket对数据库连接的获取与释放

app/concrete.py

@@ -56,20 +56,18 @@ class XF_ASR(ASR):
             await self.websocket.send(make_continue_frame(audio_data))
         elif self.status == LAST_FRAME: #发送最后一帧
             await self.websocket.send(make_last_frame(audio_data))
+            logger.debug("发送完毕")
             self.current_message += parse_xfasr_recv(json.loads(await self.websocket.recv()))
             if self.current_message == "":
                 raise AsrResultNoneError()
-            if "进入沉默模式" in self.current_message:
+            if self.current_message in ["啊"]:
+                raise SideNoiseError()
+            if "闭嘴" in self.current_message:
                 self.is_slience = True
                 asyncio.create_task(self.websocket.close())
                 raise EnterSlienceMode()
-            if "退出沉默模式" in self.current_message:
-                self.is_slience = False
-                self.current_message = "已退出沉默模式"
-            if self.is_slience:
-                asyncio.create_task(self.websocket.close())
-                raise SlienceMode()
             asyncio.create_task(self.websocket.close())
+            logger.debug(f"ASR结果: {self.current_message}")
             return [{"text":self.current_message, "audio":self.audio}]
         
         current_time = asyncio.get_event_loop().time()
@@ -86,7 +84,7 @@ class MINIMAX_LLM(LLM):
     def __init__(self):
         self.token = 0
     
-    async def chat(self, assistant, prompt, db):
+    async def chat(self, assistant, prompt):
         llm_info = json.loads(assistant.llm_info)
         messages = json.loads(assistant.messages)
         messages.append({'role':'user','content':prompt})
@@ -111,12 +109,10 @@ class MINIMAX_LLM(LLM):
                         yield msg_frame
                     except LLMResponseEnd:
                         msg_frame = {"is_end":True,"code":200,"msg":""}
+                        assistant.token = self.token
                         if self.token > llm_info['max_tokens'] * 0.8:  #如果token超过80%，则重置session
                             msg_frame['code'] = '201'
-                            as_query = db.query(Assistant).filter(Assistant.id == assistant.id).first()
-                            as_query.messages = json.dumps([{'role':'system','content':assistant.system_prompt},{'role':'user','content':prompt}])
-                            db.commit()
-                            assistant.messages = as_query.messages
+                            assistant.messages = json.dumps([{'role':'system','content':assistant.system_prompt},{'role':'user','content':prompt}])
                         yield msg_frame
 
                        
@@ -147,7 +143,7 @@ class VOLCENGINE_LLM(LLM):
         self.token = 0
         self.client = Ark(api_key=Config.VOLCENGINE_LLM.API_KEY)
     
-    async def chat(self, assistant, prompt, db):
+    async def chat(self, assistant, prompt):
         llm_info = json.loads(assistant.llm_info)
         model = self.__get_model(llm_info)
         messages = json.loads(assistant.messages)
@@ -167,18 +163,20 @@ class VOLCENGINE_LLM(LLM):
                 msg_frame = {"is_end":False,"code":200,"msg":chunk_msg}
                 yield msg_frame
             except LLMResponseEnd:
-                msg_frame = {"is_end":True,"code":20-0,"msg":""}
+                msg_frame = {"is_end":True,"code":200,"msg":""}
+                assistant.token = self.token
                 if self.token > llm_info['max_tokens'] * 0.8:  #如果token超过80%，则重置session
                     msg_frame['code'] = '201'
-                    as_query = db.query(Assistant).filter(Assistant.id == assistant.id).first()
-                    as_query.messages = json.dumps([{'role':'system','content':assistant.system_prompt},{'role':'user','content':prompt}])
-                    db.commit()
-                    assistant.messages = as_query.messages
+                    assistant.messages = json.dumps([{'role':'system','content':assistant.system_prompt},{'role':'user','content':prompt}])
                 yield msg_frame
 
     def __get_model(self, llm_info):
         if llm_info['model'] == 'doubao-4k-lite':
             return Config.VOLCENGINE_LLM.DOUBAO_LITE_4k
+        elif llm_info['model'] == 'doubao-32k-lite':
+            return Config.VOLCENGINE_LLM.DOUBAO_LITE_32k
+        elif llm_info['model'] == 'doubao-32k-pro':
+            return Config.VOLCENGINE_LLM.DOUBAO_PRO_32k
         else:
             raise UnknownVolcEngineModelError()
     
@@ -322,31 +320,31 @@ class Agent():
         self.recorder = Recorder(user_id)
     
     # 对用户输入的音频进行预处理
-    def user_audio_process(self, audio, db):
+    def user_audio_process(self, audio):
         return self.user_audio_service_chain.user_audio_process(audio, recorder=self.recorder)
     
     # 进行流式语音识别
-    async def stream_recognize(self, chunk, db):
+    async def stream_recognize(self, chunk):
         return await self.asr.stream_recognize(chunk)
     
     # 进行Prompt加工
-    def prompt_process(self, asr_results, db):
+    def prompt_process(self, asr_results):
         return self.prompt_service_chain.prompt_process(asr_results)
     
     # 进行大模型调用
-    async def chat(self, assistant ,prompt, db):
-        return self.llm.chat(assistant, prompt, db)
+    async def chat(self, assistant ,prompt):
+        return self.llm.chat(assistant, prompt)
     
     # 对大模型的返回进行处理
-    def llm_msg_process(self, llm_chunk, db):
+    def llm_msg_process(self, llm_chunk):
         return self.llm_msg_service_chain.llm_msg_process(llm_chunk)
     
     # 进行TTS合成
-    def synthetize(self, assistant, text, db):
+    def synthetize(self, assistant, text):
         return self.tts.synthetize(assistant, text)
     
     # 对合成后的音频进行处理
-    def tts_audio_process(self, audio, db):
+    def tts_audio_process(self, audio):
         return self.tts_audio_service_chain.tts_audio_process(audio, recorder=self.recorder)
     
     # 编码