修复了语音识别功能

src/chat/utils/utils_voice.py

@@ -15,13 +15,8 @@ async def get_voice_text(voice_base64: str) -> str:
         logger.warning("语音识别未启用，无法处理语音消息")
         return "[语音]"
     try:
-        # 解码base64音频数据
-        # 确保base64字符串只包含ASCII字符
-        if isinstance(voice_base64, str):
-            voice_base64 = voice_base64.encode("ascii", errors="ignore").decode("ascii")
-        voice_bytes = base64.b64decode(voice_base64)
         _llm = LLMRequest(model_set=model_config.model_task_config.voice, request_type="voice")
-        text = await _llm.generate_response_for_voice(voice_bytes)
+        text = await _llm.generate_response_for_voice(voice_base64)
         if text is None:
             logger.warning("未能生成语音文本")
             return "[语音(文本生成失败)]"

src/llm_models/model_client/base_client.py

@@ -113,6 +113,20 @@ class BaseClient:
         :return: 嵌入响应
         """
         raise RuntimeError("This method should be overridden in subclasses")
+    
+    async def get_audio_transcriptions(
+        self,
+        model_info: ModelInfo,
+        message_list: list[Message],
+        extra_params: dict[str, Any] | None = None,
+    ) -> APIResponse:
+        """
+        获取音频转录
+        :param model_info: 模型信息
+        :param message_list: 消息列表，包含音频内容
+        :return: 音频转录响应
+        """
+        raise RuntimeError("This method should be overridden in subclasses")
 
 
 class ClientRegistry:

src/llm_models/model_client/openai_client.py

@@ -532,3 +532,37 @@ class OpenaiClient(BaseClient):
             )
 
         return response
+
+    async def get_audio_transcriptions(
+        self,
+        model_info: ModelInfo,
+        message_list: list[Message],
+        extra_params: dict[str, Any] | None = None,
+    ) -> APIResponse:
+        """
+        获取音频转录
+        :param model_info: 模型信息
+        :param audio_base64: 音频的base64编码
+        :return: 转录响应
+        """
+        try:
+            raw_response = await self.client.audio.transcriptions.create(
+                model=model_info.model_identifier,
+                file=message_list[0].content[0],
+                extra_body=extra_params
+            )
+        except APIConnectionError as e:
+            raise NetworkConnectionError() from e
+        except APIStatusError as e:
+            # 重封装APIError为RespNotOkException
+            raise RespNotOkException(e.status_code) from e
+        response = APIResponse()
+        # 解析转录响应
+        if hasattr(raw_response, "text"):
+            response.content = raw_response.text
+        else:
+            raise RespParseException(
+                raw_response,
+                "响应解析失败，缺失转录文本。",
+            )
+        return response

src/llm_models/payload_content/message.py

@@ -1,5 +1,6 @@
+import base64
 from enum import Enum
-
+from io import BytesIO
 
 # 设计这系列类的目的是为未来可能的扩展做准备
 
@@ -54,6 +55,20 @@ class MessageBuilder:
         self.__content.append(text)
         return self
 
+    def add_file_content(
+        self, file_name: str, file_base64: str
+    ) -> "MessageBuilder":
+        """
+        添加文件内容
+        :param file_name: 文件名(包含类型后缀)
+        :param file_base64: 文件的base64编码
+        :return: MessageBuilder对象
+        """
+        if not file_name or not file_base64:
+            raise ValueError("文件名和base64编码不能为空")
+        self.__content.append((file_name, BytesIO(base64.b64decode(file_base64))))
+        return self
+    
     def add_image_content(
         self, image_format: str, image_base64: str
     ) -> "MessageBuilder":

src/llm_models/utils_model.py

@@ -38,7 +38,7 @@ class RequestType(Enum):
 
     RESPONSE = "response"
     EMBEDDING = "embedding"
-
+    AUDIO = "audio"
 
 class LLMRequest:
     """LLM请求类"""
@@ -106,8 +106,32 @@ class LLMRequest:
             )
         return content, (reasoning_content, model_info.name, tool_calls)
 
-    async def generate_response_for_voice(self):
-        pass
+    async def generate_response_for_voice(self, voice_base64: str) -> Optional[str]:
+        """
+        为语音生成响应
+        Args:
+            voice_base64 (str): 语音的Base64编码字符串
+        Returns:
+            (Optional[str]): 生成的文本描述或None
+        """
+        # 请求体构建
+        message_builder = MessageBuilder()
+        message_builder.add_file_content(file_name="audio.wav", file_base64=voice_base64)
+        messages = [message_builder.build()]
+
+        # 模型选择
+        model_info, api_provider, client = self._select_model()
+
+        # 请求并处理返回值
+        response = await self._execute_request(
+            api_provider=api_provider,
+            client=client,
+            request_type=RequestType.AUDIO,
+            model_info=model_info,
+            message_list=messages,
+        )
+        return response.content or None
+
 
     async def generate_response_async(
         self,
@@ -255,6 +279,13 @@ class LLMRequest:
                         embedding_input=embedding_input,
                         extra_params=model_info.extra_params,
                     )
+                elif request_type == RequestType.AUDIO:
+                    assert message_list is not None, "message_list cannot be None for audio requests"
+                    return await client.get_audio_transcriptions(
+                        model_info=model_info,
+                        message_list=message_list,
+                        extra_params=model_info.extra_params,
+                    )
             except Exception as e:
                 logger.debug(f"请求失败: {str(e)}")
                 # 处理异常