debug: 修复了多个bug

1.完成了端侧发送杂音识别
2.修复minimax返回帧解析错误bug
3.修复了返回过慢的bug

app/concrete.py

@@ -2,6 +2,8 @@ from utils.xf_asr_utils import xf_asr_websocket_factory, make_first_frame, make_
 from .model import Assistant
 from .abstract import *
 from .public import *
+from .exception import *
+from .dependency import get_logger
 from utils.vits_utils import TextToSpeech
 from config import Config
 import aiohttp
@@ -14,6 +16,9 @@ import json
 vits = TextToSpeech()
 # ---------------------------------- #
 
+# ---------- 初始化logger ---------- #
+logger = get_logger()
+# ---------------------------------- #
 
 
 #------ 具体 ASR, LLM, TTS 类 ------ #
@@ -31,6 +36,8 @@ class XF_ASR(ASR):
         self.segment_start_time = None
     
     async def stream_recognize(self, chunk):
+        if self.status == FIRST_FRAME and chunk['meta_info']['is_end']: #如果是第一帧，且为end，则判断为杂音
+            raise SideNoiseError()
         if self.websocket is None: #如果websocket未建立，则建立一个新的连接
             self.websocket = await xf_asr_websocket_factory()
         if self.segment_start_time is None: #如果是第一段，则记录开始时间
@@ -50,8 +57,7 @@ class XF_ASR(ASR):
             self.current_message += parse_xfasr_recv(json.loads(await self.websocket.recv()))
             if self.current_message == "":
                 raise AsrResultNoneError()
-            await self.websocket.close()
-            print("语音识别结束，用户消息：", self.current_message)
+            asyncio.create_task(self.websocket.close())
             return [{"text":self.current_message, "audio":self.audio}]
         
         current_time = asyncio.get_event_loop().time()
@@ -71,8 +77,7 @@ class MINIMAX_LLM(LLM):
     async def chat(self, assistant, prompt, db):
         llm_info = json.loads(assistant.llm_info)
         messages = json.loads(assistant.messages)
-        messages.append({"role":"user","content":prompt})
-        assistant.messages = json.dumps(messages)
+        messages.append({'role':'user','content':prompt})
         payload = json.dumps({
             "model": llm_info['model'],
             "stream": True,
@@ -104,17 +109,26 @@ class MINIMAX_LLM(LLM):
 
                        
     def __parseChunk(self, llm_chunk):
-        result = ""
-        data=json.loads(llm_chunk.decode('utf-8')[6:])
-        if data["object"] == "chat.completion": #如果是结束帧
-            self.token = data['usage']['total_tokens']
-            raise LLMResponseEnd()
-        elif data['object'] == 'chat.completion.chunk':
-            for choice in data['choices']:
-                result += choice['delta']['content']
-        else:
-            raise UnkownLLMFrame()
-        return result
+        try:
+            result = ""
+            chunk_decoded = llm_chunk.decode('utf-8')
+            chunks = chunk_decoded.split('\n\n')
+            for chunk in chunks:
+                if not chunk:
+                    continue
+                data=json.loads(chunk[6:])
+                if data["object"] == "chat.completion": #如果是结束帧
+                    self.token = data['usage']['total_tokens']
+                    raise LLMResponseEnd()
+                elif data['object'] == 'chat.completion.chunk':
+                    for choice in data['choices']:
+                        result += choice['delta']['content']
+                else:
+                    raise UnkownLLMFrame()
+            return result
+        except (json.JSONDecodeError, KeyError):
+            logger.error(llm_chunk)
+            raise AbnormalLLMFrame(f"error llm_chunk:{llm_chunk}")
 
 class VITS_TTS(TTS):
     def __init__(self):