fix:function call bug (#268)

* 优化function call消息处理

* fix:用户说话重复记录bug

* 2025-3-10-优化llm intent (#254)

Co-authored-by: 欣南科技 <huangrongzhuang@xin-nan.com>

* update:回复上版提示词，无需再二次识别歌曲名

---------

Co-authored-by: 玄凤科技 <eric230308@gmail.com>
Co-authored-by: hrz <1710360675@qq.com>
Co-authored-by: Jiao Haoyang <108573524+XuSenfeng@users.noreply.github.com>

docs/Deployment.md

@@ -93,7 +93,7 @@ docker logs -f xiaozhi-esp32-server
 ```
 docker stop xiaozhi-esp32-server
 docker rm xiaozhi-esp32-server
-docker rmi ghcr.nju.edu.cn/xinnan-tech/xiaozhi-esp32-server:latest
+docker rmi ghcr.nju.edu.cn/xinnan-tech/xiaozhi-esp32-server:server_latest
 ```
 
 3、重新按docker方式部署
@@ -299,14 +299,14 @@ LLM:
 
 以下是一些常见问题，供参考：
 
-[1、为什么我说的话，小智识别出来很多韩文、日文、英文](../README.md#1tts-%E7%BB%8F%E5%B8%B8%E5%A4%B1%E8%B4%A5%E7%BB%8F%E5%B8%B8%E8%B6%85%E6%97%B6-)
+[1、为什么我说的话，小智识别出来很多韩文、日文、英文](../README.md#1%E4%B8%BA%E4%BB%80%E4%B9%88%E6%88%91%E8%AF%B4%E7%9A%84%E8%AF%9D%E5%B0%8F%E6%99%BA%E8%AF%86%E5%88%AB%E5%87%BA%E6%9D%A5%E5%BE%88%E5%A4%9A%E9%9F%A9%E6%96%87%E6%97%A5%E6%96%87%E8%8B%B1%E6%96%87)
 
-[2、为什么会出现“TTS 任务出错 文件不存在”？](../README.md#1tts-%E7%BB%8F%E5%B8%B8%E5%A4%B1%E8%B4%A5%E7%BB%8F%E5%B8%B8%E8%B6%85%E6%97%B6-)
+[2、为什么会出现“TTS 任务出错 文件不存在”？](../README.md#1%E4%B8%BA%E4%BB%80%E4%B9%88%E6%88%91%E8%AF%B4%E7%9A%84%E8%AF%9D%E5%B0%8F%E6%99%BA%E8%AF%86%E5%88%AB%E5%87%BA%E6%9D%A5%E5%BE%88%E5%A4%9A%E9%9F%A9%E6%96%87%E6%97%A5%E6%96%87%E8%8B%B1%E6%96%87)
 
-[3、TTS 经常失败，经常超时](../README.md#1tts-%E7%BB%8F%E5%B8%B8%E5%A4%B1%E8%B4%A5%E7%BB%8F%E5%B8%B8%E8%B6%85%E6%97%B6-)
+[3、TTS 经常失败，经常超时](../README.md#1%E4%B8%BA%E4%BB%80%E4%B9%88%E6%88%91%E8%AF%B4%E7%9A%84%E8%AF%9D%E5%B0%8F%E6%99%BA%E8%AF%86%E5%88%AB%E5%87%BA%E6%9D%A5%E5%BE%88%E5%A4%9A%E9%9F%A9%E6%96%87%E6%97%A5%E6%96%87%E8%8B%B1%E6%96%87)
 
-[4、如何提高小智对话响应速度？](../README.md#1tts-%E7%BB%8F%E5%B8%B8%E5%A4%B1%E8%B4%A5%E7%BB%8F%E5%B8%B8%E8%B6%85%E6%97%B6-)
+[4、如何提高小智对话响应速度？](../README.md#1%E4%B8%BA%E4%BB%80%E4%B9%88%E6%88%91%E8%AF%B4%E7%9A%84%E8%AF%9D%E5%B0%8F%E6%99%BA%E8%AF%86%E5%88%AB%E5%87%BA%E6%9D%A5%E5%BE%88%E5%A4%9A%E9%9F%A9%E6%96%87%E6%97%A5%E6%96%87%E8%8B%B1%E6%96%87)
 
-[5、我说话很慢，停顿时小智老是抢话](../README.md#1tts-%E7%BB%8F%E5%B8%B8%E5%A4%B1%E8%B4%A5%E7%BB%8F%E5%B8%B8%E8%B6%85%E6%97%B6-)
+[5、我说话很慢，停顿时小智老是抢话](../README.md#1%E4%B8%BA%E4%BB%80%E4%B9%88%E6%88%91%E8%AF%B4%E7%9A%84%E8%AF%9D%E5%B0%8F%E6%99%BA%E8%AF%86%E5%88%AB%E5%87%BA%E6%9D%A5%E5%BE%88%E5%A4%9A%E9%9F%A9%E6%96%87%E6%97%A5%E6%96%87%E8%8B%B1%E6%96%87)
 
-[6、我想通过小智控制电灯、空调、远程开关机等操作](../README.md#1tts-%E7%BB%8F%E5%B8%B8%E5%A4%B1%E8%B4%A5%E7%BB%8F%E5%B8%B8%E8%B6%85%E6%97%B6-)
+[6、我想通过小智控制电灯、空调、远程开关机等操作](../README.md#1%E4%B8%BA%E4%BB%80%E4%B9%88%E6%88%91%E8%AF%B4%E7%9A%84%E8%AF%9D%E5%B0%8F%E6%99%BA%E8%AF%86%E5%88%AB%E5%87%BA%E6%9D%A5%E5%BE%88%E5%A4%9A%E9%9F%A9%E6%96%87%E6%97%A5%E6%96%87%E8%8B%B1%E6%96%87)

main/manager-web/src/apis/module/user.js

@@ -43,4 +43,19 @@ export default {
                 })
             }).send()
     },
+    // 解绑设备
+    unbindDevice(device_id, callback) {
+        RequestService.sendRequest()
+            .url(`${getServiceUrl()}/api/v1/user/device/unbind/${device_id}`)
+            .method('PUT')
+            .success((res) => {
+                RequestService.clearRequestTime();
+                callback(res);
+            })
+            .fail(() => {
+                RequestService.reAjaxFun(() => {
+                  this.unbindDevice(device_id, callback);
+                });
+              }).send()
+    },
 }

main/manager-web/src/utils/constant.js

@@ -1,6 +1,9 @@
 const HAVE_NO_RESULT = '暂无'
 export default {
     HAVE_NO_RESULT, // 项目的配置信息
+    PAGE: {
+        LOGIN: '/login',
+    },
     STORAGE_KEY: {
         TOKEN: 'TOKEN',
         PUBLIC_KEY: 'PUBLIC_KEY',

main/manager-web/src/views/home.vue

@@ -21,9 +21,9 @@
           </div>
           <div style="display: flex;align-items: center;gap: 8px;margin-top: 10px">
             <div class="serach-box">
-              <el-input placeholder="输入名称搜索.." v-model="serach" style="border: none; background: transparent;" />
+              <el-input placeholder="输入名称搜索.." v-model="serach" style="border: none; background: transparent;" @keyup.enter.native="handleSearch"  />
               <img src="@/assets/home/search.png" alt=""
-                style="width: 12px;height: 12px;margin-right: 11px;cursor: pointer;" />
+                style="width: 12px;height: 12px;margin-right: 11px;cursor: pointer;" @click="handleSearch" />
             </div>
             <img src="@/assets/home/avatar.png" alt="" style="width: 21px;height: 21px;" />
             <div class="user-info">
@@ -56,16 +56,16 @@
             </div>
           </div>
           <div
-            style="display: flex;flex-wrap: wrap;margin-top: 15px;gap: 15px;justify-content: space-between;box-sizing: border-box;">
+            style="display: flex;flex-wrap: wrap;margin-top: 15px;gap: 15px;justify-content: flex-start;box-sizing: border-box;">
-            <div class="device-item" v-for="(item,index) in deviceList" :key="index">
+            <div class="device-item" v-for="(item,index) in filteredDeviceList" :key="index">
-              <div style="display: flex;justify-content: space-between;">
+              <div style="display: flex;justify-content: space-between; align-items: center; ">
                 <div style="font-weight: 700;font-size: 18px;text-align: left;color: #3d4566;">
 <!--                  CC:ba:97:11:a6:ac-->
                   {{item.list[0]?.mac_address}}
                 </div>
-                <div>
+                <div style="display: flex;align-items: center;">
                   <img src="@/assets/home/delete.png" alt=""
-                    style="width: 18px;height: 18px;margin-right: 8px;" />
+                    style="width: 18px;height: 18px;margin-right: 8px;" @click="unbindDevice(item.list[0]?.id)" />
                   <img src="@/assets/home/info.png" alt="" style="width: 18px;height: 18px;" />
                 </div>
               </div>
@@ -229,7 +229,9 @@ export default {
   name: 'home',
   data() {
     return {
-      serach: '',
+      serach: '', // 搜索框输入内容
+      deviceList: [], // 原始设备列表
+      filteredDeviceList: [], // 过滤后的设备列表
       switchValue: false,
       addDeviceDialogVisible: false,
       deviceCode: "",
@@ -250,8 +252,7 @@ export default {
       }],
       userInfo: {
         mobile: '' // 初始化用户信息
-      },
+      }
-      deviceList:[]
     };
   },
   methods: {
@@ -270,14 +271,51 @@ export default {
     // 获取已绑设备
     getList(){
       Api.user.getHomeList(({data})=>{
-        console.log(data.data)
+        this.deviceList = data.data; // 保存原始设备列表
-        this.deviceList = data.data
+        this.filteredDeviceList = data.data; // 初始化过滤后的设备列表
       })
-    }
+    },
+    // 处理搜索
+    handleSearch() {
+      if (this.serach.trim() === '') {
+        // 如果搜索框为空，显示全部设备
+        this.filteredDeviceList = this.deviceList;
+      } else {
+        // 过滤设备列表
+        this.filteredDeviceList = this.deviceList.filter(device => {
+          return (
+            device.list[0]?.mac_address?.includes(this.serach) ||  // 匹配MAC地址
+            device.list[0]?.device_type?.includes(this.serach) ||  // 匹配设备型号
+            device.list[0]?.app_version?.includes(this.serach)     // 匹配APP版本
+          );
+        });
+      }
+    },
+    // 解绑设备
+    unbindDevice(device_id) {
+      this.$confirm('确定要解绑该设备吗？', '提示', {
+        confirmButtonText: '确定',
+        cancelButtonText: '取消',
+        type: 'warning'
+      }).then(() => {
+        // 调用解绑设备的接口
+        Api.user.unbindDevice(device_id, ({ data }) => {
+          if (data.code === 0) {
+            this.$message.success('解绑成功');
+            this.getList();
+          } else {
+            this.$message.error(data.msg || '解绑失败');
+          }
+        });
+      }).catch(() => {
+        // 用户取消操作
+        this.$message.info('已取消解绑');
+      });
+    },
   },
   mounted() {
     this.fetchUserInfo(); // 组件加载时获取用户信息
-    this.getList()
+    this.getList(); // 初始化设备列表
   }
 }
 </script>

main/xiaozhi-server/core/connection.py

@@ -11,7 +11,7 @@ import websockets
 from typing import Dict, Any
 from core.utils.dialogue import Message, Dialogue
 from core.handle.textHandle import handleTextMessage
-from core.utils.util import get_string_no_punctuation_or_emoji
+from core.utils.util import get_string_no_punctuation_or_emoji, extract_json_from_string
 from concurrent.futures import ThreadPoolExecutor, TimeoutError
 from core.handle.sendAudioHandle import sendAudioMessage
 from core.handle.receiveAudioHandle import handleAudioMessage
@@ -307,7 +307,6 @@ class ConnectionHandler:
 
         response_message = []
         processed_chars = 0  # 跟踪已处理的字符位置
-        function_call_data = None  # 存储function call数据
    
         try:
             start_time = time.time()
@@ -316,7 +315,7 @@ class ConnectionHandler:
             future = asyncio.run_coroutine_threadsafe(self.memory.query_memory(query), self.loop)
             memory_str = future.result()
 
-            # self.logger.bind(tag=TAG).info(f"记忆内容: {memory_str}")
+            #self.logger.bind(tag=TAG).info(f"对话记录: {self.dialogue.get_llm_dialogue_with_memory(memory_str)}")
 
             # 使用支持functions的streaming接口
             llm_responses = self.llm.response_with_functions(
@@ -332,48 +331,61 @@ class ConnectionHandler:
         text_index = 0
 
         # 处理流式响应
+        tool_call_flag = False
+        function_name = None
+        function_id = None
+        function_arguments = ""
+        content_arguments = ""
         for response in llm_responses:
-            if response["type"] == "content":
+            content, tools_call = response
-                content = response["content"]
+            if content is not None and len(content)>0:
-                response_message.append(content)
+                if len(response_message)<=0 and content=="```":
+                    tool_call_flag = True
 
-                if self.client_abort:
+            if tools_call is not None:
-                    break
+                tool_call_flag = True
+                if tools_call[0].id is not None:
+                    function_id = tools_call[0].id
+                if tools_call[0].function.name is not None:
+                    function_name = tools_call[0].function.name
+                if tools_call[0].function.arguments is not None:
+                    function_arguments += tools_call[0].function.arguments
 
-                end_time = time.time()
+            if content is not None and len(content) > 0:
-                self.logger.bind(tag=TAG).debug(f"大模型返回时间: {end_time - start_time} 秒, 生成token={content}")
+                if tool_call_flag:
+                    content_arguments+=content
+                else:
+                    response_message.append(content)
 
-                # 处理文本分段和TTS逻辑
+                    if self.client_abort:
-                # 合并当前全部文本并处理未分割部分
+                        break
-                full_text = "".join(response_message)
-                current_text = full_text[processed_chars:]  # 从未处理的位置开始
 
-                # 查找最后一个有效标点
+                    end_time = time.time()
-                punctuations = ("。", "？", "！", "；", "：")
+                    self.logger.bind(tag=TAG).debug(f"大模型返回时间: {end_time - start_time} 秒, 生成token={content}")
-                last_punct_pos = -1
-                for punct in punctuations:
-                    pos = current_text.rfind(punct)
-                    if pos > last_punct_pos:
-                        last_punct_pos = pos
 
-                # 找到分割点则处理
+                    # 处理文本分段和TTS逻辑
-                if last_punct_pos != -1:
+                    # 合并当前全部文本并处理未分割部分
-                    segment_text_raw = current_text[:last_punct_pos + 1]
+                    full_text = "".join(response_message)
-                    segment_text = get_string_no_punctuation_or_emoji(segment_text_raw)
+                    current_text = full_text[processed_chars:]  # 从未处理的位置开始
-                    if segment_text:
-                        text_index += 1
-                        self.recode_first_last_text(segment_text, text_index)
-                        future = self.executor.submit(self.speak_and_play, segment_text, text_index)
-                        self.tts_queue.put(future)
-                        processed_chars += len(segment_text_raw)  # 更新已处理字符位置
 
-            elif response["type"] == "function_call":
+                    # 查找最后一个有效标点
-                # Extract function call data
+                    punctuations = ("。", "？", "！", "；", "：")
-                function_call_data = {
+                    last_punct_pos = -1
-                    "name": response["function_call"]["function"]["name"],
+                    for punct in punctuations:
-                    "arguments": response["function_call"]["function"]["arguments"]
+                        pos = current_text.rfind(punct)
-                }
+                        if pos > last_punct_pos:
-                self.logger.bind(tag=TAG).info(f"Function call detected: {function_call_data}")
+                            last_punct_pos = pos
+
+                    # 找到分割点则处理
+                    if last_punct_pos != -1:
+                        segment_text_raw = current_text[:last_punct_pos + 1]
+                        segment_text = get_string_no_punctuation_or_emoji(segment_text_raw)
+                        if segment_text:
+                            text_index += 1
+                            self.recode_first_last_text(segment_text, text_index)
+                            future = self.executor.submit(self.speak_and_play, segment_text, text_index)
+                            self.tts_queue.put(future)
+                            processed_chars += len(segment_text_raw)  # 更新已处理字符位置
 
         # 处理最后剩余的文本
         full_text = "".join(response_message)
@@ -387,23 +399,46 @@ class ConnectionHandler:
                 self.tts_queue.put(future)
 
         # 存储对话内容
-        self.dialogue.put(Message(role="assistant", content="".join(response_message)))
+        if len(response_message)>0:
+            self.dialogue.put(Message(role="assistant", content="".join(response_message)))
 
         # 处理function call
-        if function_call_data:
+        if tool_call_flag:
+            if function_id is None:
+                a = extract_json_from_string(content_arguments)
+                if a is not None:
+                    content_arguments_json = json.loads(a)
+                    function_name = content_arguments_json["function_name"]
+                    function_arguments = json.dumps(content_arguments_json["args"], ensure_ascii=False)
+                    function_id = str(uuid.uuid4().hex)
+                else:
+                    return []
+                function_arguments = json.loads(function_arguments)
+            self.logger.bind(tag=TAG).info(f"function_name={function_name}, function_id={function_id}, function_arguments={function_arguments}")
+            function_call_data = {
+                "name": function_name,
+                "id": function_id,
+                "arguments": function_arguments
+            }
             result = handle_llm_function_call(self, function_call_data)
-            if result.action == Action.RESPONSE:
+            self._handle_function_result(result, function_call_data, text_index+1)
-                text = result.response
-                text_index += 1
-                self.recode_first_last_text(text, text_index)
-                future = self.executor.submit(self.speak_and_play, text, text_index)
-                self.tts_queue.put(future)
 
         self.llm_finish_task = True
         self.logger.bind(tag=TAG).debug(json.dumps(self.dialogue.get_llm_dialogue(), indent=4, ensure_ascii=False))
 
         return True
 
+    def _handle_function_result(self, result, function_call_data, text_index):
+        if result.action == Action.RESPONSE: # 直接回复前端
+            text = result.response
+            self.recode_first_last_text(text, text_index)
+            future = self.executor.submit(self.speak_and_play, text, text_index)
+            self.tts_queue.put(future)
+            self.dialogue.put(Message(role="assistant", content=text))
+        if result.action == Action.REQLLM: # 调用函数后再请求llm生成回复
+            text = result.response
+        
+
     def _tts_priority_thread(self):
         while not self.stop_event.is_set():
             text = None

main/xiaozhi-server/core/handle/intentHandler.py

@@ -122,12 +122,10 @@ async def analyze_intent_with_llm(conn, text):
         logger.bind(tag=TAG).warning("意图识别服务未初始化")
         return None
 
-    # 创建对话历史记录
+    # 对话历史记录
     dialogue = conn.dialogue
-    dialogue.put(Message(role="user", content=text))
-
     try:
-        intent_result = await conn.intent.detect_intent(dialogue.dialogue)
+        intent_result = await conn.intent.detect_intent(dialogue.dialogue, text)
         logger.bind(tag=TAG).info(f"意图识别结果: {intent_result}")
 
         # 尝试解析JSON结果

main/xiaozhi-server/core/handle/receiveAudioHandle.py

@@ -67,8 +67,9 @@ async def no_voice_close_connect(conn):
     else:
         no_voice_time = time.time() * 1000 - conn.client_no_voice_last_time
         close_connection_no_voice_time = conn.config.get("close_connection_no_voice_time", 120)
-        if no_voice_time > 1000 * close_connection_no_voice_time:
+        if not conn.close_after_chat and no_voice_time > 1000 * close_connection_no_voice_time:
+            conn.close_after_chat = True
             conn.client_abort = False
             conn.asr_server_receive = False
-            prompt = "时间过得真快，我都好久没说话了。请你用十个字左右话跟我告别，以“再见”或“拜拜”为结尾"
+            prompt = "请你以“时间过得真快”未来头，用富有感情、依依不舍的话来结束这场对话吧。"
             await startToChat(conn, prompt)

main/xiaozhi-server/core/handle/sendAudioHandle.py

@@ -13,28 +13,37 @@ async def sendAudioMessage(conn, audios, text, text_index=0):
         logger.bind(tag=TAG).info(f"发送第一段语音: {text}")
     await send_tts_message(conn, "sentence_start", text)
 
-    # 初始化流控参数
+    # 流控参数优化
-    frame_duration = 60  # 毫秒
+    original_frame_duration = 60  # 原始帧时长（毫秒）
-    start_time = time.perf_counter()  # 使用高精度计时器
+    adjusted_frame_duration = int(original_frame_duration * 0.8)  # 缩短20%
-    play_position = 0  # 已播放的时长（毫秒）
+    total_frames = len(audios)  # 获取总帧数
+    compensation = total_frames * (original_frame_duration - adjusted_frame_duration) / 1000  # 补偿时间（秒）
+
+    start_time = time.perf_counter()
+    play_position = 0  # 已播放时长（毫秒）
 
     for opus_packet in audios:
         if conn.client_abort:
             return
 
-        # 计算当前包的预期发送时间
+        # 计算带加速因子的预期时间
         expected_time = start_time + (play_position / 1000)
         current_time = time.perf_counter()
 
-        # 等待直到预期时间
+        # 流控等待（使用加速后的帧时长）
         delay = expected_time - current_time
         if delay > 0:
             await asyncio.sleep(delay)
 
-        # 发送音频包
         await conn.websocket.send(opus_packet)
-        play_position += frame_duration  # 更新播放位置
+        play_position += adjusted_frame_duration  # 使用调整后的帧时长
+
+    # 补偿因加速损失的时长
+    if compensation > 0:
+        await asyncio.sleep(compensation)
+
     await send_tts_message(conn, "sentence_end", text)
+
     # 发送结束消息（如果是最后一个文本）
     if conn.llm_finish_task and text_index == conn.tts_last_text_index:
         await send_tts_message(conn, 'stop', None)

main/xiaozhi-server/core/providers/intent/base.py

@@ -5,6 +5,7 @@ from config.logger import setup_logging
 TAG = __name__
 logger = setup_logging()
 
+
 class IntentProviderBase(ABC):
     def __init__(self, config):
         self.config = config
@@ -19,7 +20,7 @@ class IntentProviderBase(ABC):
         logger.bind(tag=TAG).debug("Set LLM for intent provider")
 
     @abstractmethod
-    async def detect_intent(self, dialogue_history: List[Dict]) -> str:
+    async def detect_intent(self, dialogue_history: List[Dict], text: str) -> str:
         """
         检测用户最后一句话的意图
         Args:

main/xiaozhi-server/core/providers/intent/intent_llm/intent_llm.py

@@ -37,7 +37,7 @@ class IntentProvider(IntentProviderBase):
         )
         return prompt
     
-    async def detect_intent(self, dialogue_history: List[Dict]) -> str:
+    async def detect_intent(self, dialogue_history: List[Dict], text:str) -> str:
         if not self.llm:
             raise ValueError("LLM provider not set")
 
@@ -48,9 +48,8 @@ class IntentProvider(IntentProviderBase):
                 msgStr += f"User: {msg.content}\n"
             elif msg.role== "assistant":
                 msgStr += f"Assistant: {msg.content}\n"
-
+        msgStr += f"User: {text}\n"
         user_prompt = f"请分析用户的意图：\n{msgStr}"
-        
         # 使用LLM进行意图识别
         intent = self.llm.response_no_stream(
             system_prompt=self.promot,

main/xiaozhi-server/core/providers/intent/nointent/nointent.py

@@ -5,12 +5,14 @@ from config.logger import setup_logging
 TAG = __name__
 logger = setup_logging()
 
+
 class IntentProvider(IntentProviderBase):
-    async def detect_intent(self, dialogue_history: List[Dict]) -> str:
+    async def detect_intent(self, dialogue_history: List[Dict], text: str) -> str:
         """
         默认的意图识别实现，始终返回继续聊天
         Args:
             dialogue_history: 对话历史记录列表
+            text: 本次对话记录
         Returns:
             固定返回"继续聊天"
         """

main/xiaozhi-server/core/providers/llm/ollama/ollama.py

@@ -51,30 +51,8 @@ class LLMProvider(LLMProviderBase):
                 tools=functions,
             )
             
-            current_function_call = None
-            current_content = ""
-            
             for chunk in stream:
-                delta = chunk.choices[0].delta
+                yield chunk.choices[0].delta.content, chunk.choices[0].delta.tool_calls
-                
-                if delta.content:
-                    current_content += delta.content
-                    yield {"type": "content", "content": delta.content}
-                
-                if delta.tool_calls:
-                    tool_call = delta.tool_calls[0]
-                    # Handle the function call data using proper attribute access
-                    if not current_function_call:
-                        current_function_call = {
-                            "function": {
-                                "name": tool_call.function.name,
-                                "arguments": tool_call.function.arguments
-                            }
-                        }
-            
-            if current_function_call:
-                logger.bind(tag=TAG).debug(f"ollama Function call detected: {current_function_call}")
-                yield {"type": "function_call", "function_call": current_function_call}
 
         except Exception as e:
             logger.bind(tag=TAG).error(f"Error in Ollama function call: {e}")

main/xiaozhi-server/core/providers/llm/openai/openai.py

@@ -53,30 +53,8 @@ class LLMProvider(LLMProviderBase):
                 tools=functions,
             )
             
-            current_function_call = None
-            current_content = ""
-            
             for chunk in stream:
-                delta = chunk.choices[0].delta
+                yield chunk.choices[0].delta.content, chunk.choices[0].delta.tool_calls
-                
-                if delta.content:
-                    current_content += delta.content
-                    yield {"type": "content", "content": delta.content}
-                
-                if delta.tool_calls:
-                    tool_call = delta.tool_calls[0]
-                    # Handle the function call data using proper attribute access
-                    if not current_function_call:
-                        current_function_call = {
-                            "function": {
-                                "name": tool_call.function.name,
-                                "arguments": tool_call.function.arguments
-                            }
-                        }
-            
-            if current_function_call:
-                logger.bind(tag=TAG).debug(f"openai Function call detected: {current_function_call}")
-                yield {"type": "function_call", "function_call": current_function_call}
                 
         except Exception as e:
             self.logger.bind(tag=TAG).error(f"Error in function call streaming: {e}")

main/xiaozhi-server/core/utils/util.py

@@ -4,6 +4,7 @@ import yaml
 import socket
 import subprocess
 import logging
+import re
 
 
 def get_project_dir():
@@ -119,3 +120,11 @@ def check_ffmpeg_installed():
         error_msg += "1、按照项目的安装文档，正确进入conda环境\n"
         error_msg += "2、查阅安装文档，如何在conda环境中安装ffmpeg\n"
         raise ValueError(error_msg)
+    
+def extract_json_from_string(input_string):
+    """提取字符串中的 JSON 部分"""
+    pattern = r'(\{.*\})'
+    match = re.search(pattern, input_string)
+    if match:
+        return match.group(1)  # 返回提取的 JSON 字符串
+    return None