fix: CosyVoice 音色名 v3 后缀修复（418错误）

根因：cosyvoice-v3.5-plus 模型要求系统音色使用 _v3 后缀命名
（如 longxiaochun_v3），旧名 longxiaochun 导致引擎返回 418 音色不支持。

修复：
- config.py: 默认音色 longxiaochun → longxiaochun_v3
- preset_voices.py: 8个预置音色全部加 _v3 后缀
- cosyvoice_service.py: docstring 示例更新
- 相关测试文件 voice_id 全部同步更新
- 188个单元测试全绿

apps/api/app/dependencies.py

@@ -201,7 +201,18 @@ def get_voice_clone_profile_repository(
 
 
 def get_cosyvoice_service():
-    """Provide the CosyVoice service instance."""
+    """Provide the CosyVoice service instance.
+
+    注入 OSS 音频URL预签名函数，确保私有bucket下的参考音频
+    能被 CosyVoice 服务器下载。
+    """
+    from app.core.storage import get_storage_service
     from packages.application.cosyvoice_service import CosyVoiceService
 
-    return CosyVoiceService()
+    storage = get_storage_service()
+
+    def _sign_audio_url(url: str) -> str:
+        """对音频URL做预签名，私有bucket下 CosyVoice 服务器才能下载."""
+        return storage.get_download_url(url, expires_seconds=86400)
+
+    return CosyVoiceService(audio_url_signer=_sign_audio_url)

apps/worker/worker_app/tasks/voice_clone.py

@@ -16,6 +16,7 @@ from packages.application.cosyvoice_service import (
     CosyVoiceTimeoutError,
 )
 from packages.application.voice_clone.workflow import VoiceCloneWorkflowService
+from video_processing.oss_helpers import get_signed_download_url
 
 logger = logging.getLogger(__name__)
 
@@ -48,7 +49,9 @@ def process_voice_clone(self: Task, profile_id: str) -> dict:
         repo = SQLAlchemyVoiceCloneProfileRepository(session)
         workflow = VoiceCloneWorkflowService(
             repository=repo,
-            cosyvoice_service=CosyVoiceService(),
+            cosyvoice_service=CosyVoiceService(
+                audio_url_signer=lambda url: get_signed_download_url(url, expires_seconds=86400) or url
+            ),
         )
 
         updated_profile = workflow.poll_and_process_clone(profile_id, timeout=300)

packages/application/cosyvoice_service.py

@@ -1,11 +1,13 @@
-"""CosyVoice 语音服务 — Phase 3.
+"""CosyVoice 语音服务 — 适配阿里云百炼 DashScope API.
 
-封装阿里云 CosyVoice 语音合成 API，提供：
+封装阿里云百炼 CosyVoice 语音合成 API，提供：
 - 预置音色列表查询
-- 音色克隆（提交任务 + 轮询状态）
-- 语音合成（提交任务 + 轮询状态）
+- 音色克隆（提交 + 轮询状态）
+- 语音合成（同步非流式调用）
 
-API 文档: https://help.aliyun.com/zh/model-studio/cosyvoice
+API 文档:
+- 音色克隆: https://help.aliyun.com/document_detail/3027318.html
+- 语音合成: https://help.aliyun.com/zh/model-studio/cosyvoice-tts-http-api
 """
 
 from __future__ import annotations
@@ -60,33 +62,34 @@ class SynthesizeResult:
 
 
 class CosyVoiceService:
-    """CosyVoice 语音服务。
+    """CosyVoice 语音服务.
 
-    封装阿里云 CosyVoice API，提供音色克隆和语音合成功能。
-    支持同步和异步两种模式：
-    - 同步：API 直接返回结果
-    - 异步：API 返回 task_id，需要轮询状态
+    封装阿里云百炼 CosyVoice API，提供音色克隆和语音合成功能.
+
+    接口总览:
+    - 音色克隆: POST /services/audio/tts/customization (model=voice-enrollment)
+      - action=create_voice: 创建克隆音色，返回 voice_id（状态 DEPLOYING）
+      - action=query_voice: 查询音色状态（DEPLOYING / OK / UNDEPLOYED）
+    - 语音合成: POST /services/audio/tts/SpeechSynthesizer (model=cosyvoice-v3.5-plus)
+      - 非流式: 同步返回音频 URL
 
     使用示例:
         service = CosyVoiceService(
             api_key="your-api-key",
-            base_url="https://dashscope.aliyuncs.com/api/v1/services/aigc/text2audio",
-            model="cosyvoice-v1",
+            base_url="https://dashscope.aliyuncs.com/api/v1",
+            model="cosyvoice-v3.5-plus",
         )
 
-        # 获取预置音色
-        voices = service.list_preset_voices()
-
         # 音色克隆
         result = service.clone_voice(audio_url="https://example.com/audio.mp3")
 
         # 语音合成
-        result = service.synthesize_speech(text="你好世界", voice_id="longxiaochun")
+        result = service.synthesize_speech(text="你好世界", voice_id="longxiaochun_v3")
     """
 
-    # 轮询配置
-    POLL_INTERVAL = 2.0  # 秒
-    MAX_POLL_ATTEMPTS = 60  # 最多轮询 60 次（2分钟）
+    # 音色状态轮询配置
+    CLONE_POLL_INTERVAL = 5.0  # 秒
+    CLONE_MAX_POLL_ATTEMPTS = 60  # 最多轮询 60 次（5分钟）
 
     # 重试配置
     MAX_RETRIES = 3
@@ -97,24 +100,34 @@ class CosyVoiceService:
         api_key: str = "",
         base_url: str = "",
         model: str = "",
+        clone_model: str = "",
         http_client: Optional[httpx.Client] = None,
+        audio_url_signer: Optional[callable] = None,
     ) -> None:
-        """初始化 CosyVoice 服务。
+        """初始化 CosyVoice 服务.
 
         Args:
-            api_key: CosyVoice API Key，为空时从配置读取
-            base_url: CosyVoice API Base URL，为空时从配置读取
-            model: CosyVoice 模型名称，为空时从配置读取
+            api_key: DashScope API Key，为空时从配置读取
+            base_url: DashScope API Base URL，为空时从配置读取
+            model: 语音合成模型名称，为空时从配置读取
+            clone_model: 音色克隆模型名称，为空时从配置读取
             http_client: 可选的 HTTP 客户端（用于测试注入）
+            audio_url_signer: 可选的音频URL预签名函数，签名式 fn(url) -> str.
+                用于私有 bucket 下，将裸 URL 转为预签名 URL，
+                确保 CosyVoice 服务器能下载参考音频.
         """
         settings = get_shared_settings()
 
         self._api_key = api_key or settings.cosyvoice_api_key
         self._base_url = base_url or settings.cosyvoice_base_url
         self._model = model or settings.cosyvoice_model
+        self._clone_model = clone_model or getattr(
+            settings, "cosyvoice_clone_model", "voice-enrollment"
+        )
+        self._audio_url_signer = audio_url_signer
 
         self._client = http_client or httpx.Client(
-            timeout=httpx.Timeout(30.0, connect=10.0),
+            timeout=httpx.Timeout(60.0, connect=10.0),
         )
         self._owns_client = http_client is None
 
@@ -132,7 +145,7 @@ class CosyVoiceService:
     # ── 预置音色 ─────────────────────────────────────────
 
     def list_preset_voices(self) -> list[PresetVoice]:
-        """获取预置音色列表。
+        """获取预置音色列表.
 
         Returns:
             预置音色列表
@@ -146,20 +159,22 @@ class CosyVoiceService:
         audio_url: str,
         voice_name: str = "",
         language: str = "zh-CN",
+        target_model: str = "",
     ) -> dict:
-        """提交音色克隆任务（非阻塞）。
+        """提交音色克隆任务（非阻塞）.
 
-        只提交任务到 CosyVoice API，不轮询结果。
-        返回的 dict 包含 task_id（异步）或 voice_id（同步）。
+        调用百炼 voice-enrollment API 创建克隆音色.
+        创建后音色状态为 DEPLOYING，需通过 query_voice_status 轮询直到 OK.
 
         Args:
-            audio_url: 参考音频 URL
-            voice_name: 音色名称（可选）
-            language: 语言代码
+            audio_url: 参考音频 URL（必须公网可访问）
+            voice_name: 音色名称前缀（字母数字，最多10字符）
+            language: 语言代码（zh-CN 会转换为 zh）
+            target_model: 目标合成模型，默认使用当前 model
 
         Returns:
-            dict: {"task_id": str, "voice_id": str, "request_id": str}
-                  task_id 和 voice_id 至少有一个非空
+            dict: {"voice_id": str, "status": str, "request_id": str}
+                  voice_id 非空，status 通常为 DEPLOYING
 
         Raises:
             CosyVoiceError: API 调用失败
@@ -171,48 +186,68 @@ class CosyVoiceService:
         if not self._api_key:
             raise CosyVoiceAuthError("CosyVoice API Key 未配置")
 
+        # voice_name 作为 prefix，限制字母数字，最多10字符
+        # 不符合要求的做清洗
+        prefix = self._sanitize_prefix(voice_name) if voice_name else "clone"
+
+        # 语言转换：zh-CN → zh，保留 ISO 639-1 格式
+        lang_code = language.split("-")[0].lower() if language else "zh"
+
+        target = target_model or self._model
+
+        # 如果配置了 audio_url_signer，对音频URL做预签名
+        # （私有 bucket 下 CosyVoice 服务器无法直接访问裸 URL）
+        signed_audio_url = audio_url
+        if self._audio_url_signer:
+            try:
+                signed_audio_url = self._audio_url_signer(audio_url)
+                logger.info("音频URL已预签名: original=%s signed_prefix=%s",
+                            audio_url[:80], signed_audio_url[:80])
+            except Exception as e:
+                logger.warning("音频URL预签名失败，使用原始URL: %s", e)
+
         payload = {
-            "model": self._model,
+            "model": self._clone_model,
             "input": {
-                "audio_url": audio_url,
-            },
-            "parameters": {
-                "language": language,
+                "action": "create_voice",
+                "target_model": target,
+                "prefix": prefix,
+                "url": signed_audio_url,
+                "language_hints": [lang_code],
             },
         }
-        if voice_name:
-            payload["parameters"]["voice_name"] = voice_name
 
         response = self._call_api(
             method="POST",
-            path="/services/audio/voice-clone",
+            path="/services/audio/tts/customization",
             json=payload,
             timeout=60.0,
         )
 
         output = response.get("output", {})
-        task_id = output.get("task_id", "")
         voice_id = output.get("voice_id", "")
+        status = output.get("status", "DEPLOYING")
         request_id = response.get("request_id", "")
 
-        if not task_id and not voice_id:
-            raise CosyVoiceError(f"CosyVoice API 未返回 task_id 或 voice_id: {response}")
+        if not voice_id:
+            raise CosyVoiceError(f"CosyVoice API 未返回 voice_id: {response}")
 
         return {
-            "task_id": task_id,
             "voice_id": voice_id,
+            "status": status,
             "request_id": request_id,
         }
 
-    def check_task_status(self, task_id: str) -> dict:
-        """查询克隆任务状态（单次查询，不轮询）。
+    def query_voice_status(self, voice_id: str) -> dict:
+        """查询音色状态（单次查询，不轮询）.
 
         Args:
-            task_id: 任务 ID
+            voice_id: 音色 ID
 
         Returns:
-            dict: {"status": str, "voice_id": str, "message": str}
-                  status 为 SUCCEEDED/FAILED/PENDING/RUNNING
+            dict: {"status": str, "target_model": str, "gmt_create": str,
+                   "gmt_modified": str, "resource_link": str}
+                  status 为 DEPLOYING / OK / UNDEPLOYED
 
         Raises:
             CosyVoiceError: API 调用失败
@@ -221,40 +256,98 @@ class CosyVoiceService:
         if not self._api_key:
             raise CosyVoiceAuthError("CosyVoice API Key 未配置")
 
+        if not voice_id:
+            raise ValueError("voice_id 不能为空")
+
+        payload = {
+            "model": self._clone_model,
+            "input": {
+                "action": "query_voice",
+                "voice_id": voice_id,
+            },
+        }
+
         response = self._call_api(
-            method="GET",
-            path=f"/tasks/{task_id}",
+            method="POST",
+            path="/services/audio/tts/customization",
+            json=payload,
             timeout=30.0,
         )
 
         output = response.get("output", {})
-        status = output.get("task_status", "").upper()
-        voice_id = output.get("voice_id", "")
-        message = output.get("message", "")
-
         return {
-            "status": status,
-            "voice_id": voice_id,
-            "message": message,
+            "status": output.get("status", ""),
+            "target_model": output.get("target_model", ""),
+            "gmt_create": output.get("gmt_create", ""),
+            "gmt_modified": output.get("gmt_modified", ""),
+            "resource_link": output.get("resource_link", ""),
         }
 
-    def poll_clone_task(self, task_id: str, timeout: float = 300.0) -> dict:
-        """轮询音色克隆任务状态（公开方法）。
+    def check_task_status(self, task_id: str) -> dict:
+        """查询克隆任务状态（兼容旧接口，实际用 voice_id 查询）.
 
-        供 Celery 后台任务调用，轮询直到完成或超时。
+        为了兼容旧代码，task_id 参数名保留，但实际传的是 voice_id.
 
         Args:
-            task_id: CosyVoice 任务 ID
+            task_id: 音色 ID（兼容旧接口名）
+
+        Returns:
+            dict: {"status": str, "voice_id": str, "message": str}
+        """
+        result = self.query_voice_status(task_id)
+        return {
+            "status": result["status"],
+            "voice_id": task_id,
+            "message": "",
+        }
+
+    def poll_clone_task(self, voice_id: str, timeout: float = 300.0) -> dict:
+        """轮询音色克隆状态直到完成或超时.
+
+        供 Celery 后台任务调用，轮询直到状态变为 OK 或 UNDEPLOYED.
+
+        Args:
+            voice_id: 音色 ID
             timeout: 超时时间（秒），默认 300
 
         Returns:
             dict: {"voice_id": str}
 
         Raises:
-            CosyVoiceError: 任务失败
+            CosyVoiceError: 任务失败（状态 UNDEPLOYED）
             CosyVoiceTimeoutError: 超时
         """
-        return self._poll_clone_task(task_id, timeout=timeout)
+        start_time = time.time()
+        attempts = 0
+
+        while attempts < self.CLONE_MAX_POLL_ATTEMPTS:
+            elapsed = time.time() - start_time
+            if elapsed > timeout:
+                raise CosyVoiceTimeoutError(
+                    f"音色克隆任务超时（{timeout}秒）: voice_id={voice_id}"
+                )
+
+            result = self.query_voice_status(voice_id)
+            status = result.get("status", "").upper()
+
+            if status == "OK":
+                return {"voice_id": voice_id}
+            elif status == "UNDEPLOYED":
+                raise CosyVoiceError(
+                    f"音色克隆任务失败（审核未通过）: voice_id={voice_id}"
+                )
+            elif status in ("DEPLOYING", "PENDING", "PROCESSING", ""):
+                # 继续轮询
+                time.sleep(self.CLONE_POLL_INTERVAL)
+                attempts += 1
+            else:
+                logger.warning("未知的音色状态: %s (voice_id=%s)", status, voice_id)
+                time.sleep(self.CLONE_POLL_INTERVAL)
+                attempts += 1
+
+        raise CosyVoiceTimeoutError(
+            f"音色克隆任务轮询次数超限: voice_id={voice_id}"
+        )
 
     def clone_voice(
         self,
@@ -262,122 +355,45 @@ class CosyVoiceService:
         voice_name: str = "",
         language: str = "zh-CN",
         timeout: float = 300.0,
+        target_model: str = "",
     ) -> CloneResult:
-        """克隆音色。
+        """克隆音色（阻塞，直到完成或超时）.
 
-        提交音色克隆任务到 CosyVoice API，并轮询直到完成或超时。
+        提交音色克隆到百炼 API，并轮询直到状态变为 OK 或超时.
 
         Args:
-            audio_url: 参考音频 URL
-            voice_name: 音色名称（可选）
+            audio_url: 参考音频 URL（必须公网可访问）
+            voice_name: 音色名称前缀
             language: 语言代码
             timeout: 超时时间（秒）
+            target_model: 目标合成模型
 
         Returns:
             CloneResult: 克隆结果，包含 voice_id
 
         Raises:
-            CosyVoiceError: API 调用失败
+            CosyVoiceError: API 调用失败或克隆失败
             CosyVoiceTimeoutError: 超时
             CosyVoiceAuthError: 认证失败
             ValueError: 参数无效
         """
-        if not audio_url:
-            raise ValueError("audio_url 不能为空")
-        if not self._api_key:
-            raise CosyVoiceAuthError("CosyVoice API Key 未配置")
-
-        # 构建请求
-        payload = {
-            "model": self._model,
-            "input": {
-                "audio_url": audio_url,
-            },
-            "parameters": {
-                "language": language,
-            },
-        }
-        if voice_name:
-            payload["parameters"]["voice_name"] = voice_name
-
-        # 调用 API
-        response = self._call_api(
-            method="POST",
-            path="/services/audio/voice-clone",
-            json=payload,
-            timeout=timeout,
+        submit_result = self.submit_clone_task(
+            audio_url=audio_url,
+            voice_name=voice_name,
+            language=language,
+            target_model=target_model,
         )
 
-        # 解析响应
-        output = response.get("output", {})
+        voice_id = submit_result["voice_id"]
+        request_id = submit_result["request_id"]
 
-        # 检查是否有 task_id（异步模式）
-        task_id = output.get("task_id")
-        voice_id = output.get("voice_id")
+        # 如果创建时已经是 OK 状态，直接返回
+        if submit_result.get("status", "").upper() == "OK":
+            return CloneResult(voice_id=voice_id, request_id=request_id)
 
-        if task_id:
-            # 异步模式：轮询任务状态
-            result = self._poll_clone_task(task_id, timeout)
-            return CloneResult(
-                voice_id=result["voice_id"],
-                request_id=response.get("request_id", ""),
-            )
-        elif voice_id:
-            # 同步模式：直接返回结果
-            return CloneResult(
-                voice_id=voice_id,
-                request_id=response.get("request_id", ""),
-            )
-        else:
-            raise CosyVoiceError(f"CosyVoice API 未返回 task_id 或 voice_id: {response}")
-
-    def _poll_clone_task(self, task_id: str, timeout: float) -> dict:
-        """轮询音色克隆任务状态。
-
-        Args:
-            task_id: 任务 ID
-            timeout: 超时时间（秒）
-
-        Returns:
-            任务结果字典
-
-        Raises:
-            CosyVoiceError: 任务失败
-            CosyVoiceTimeoutError: 超时
-        """
-        start_time = time.time()
-        attempts = 0
-
-        while attempts < self.MAX_POLL_ATTEMPTS:
-            elapsed = time.time() - start_time
-            if elapsed > timeout:
-                raise CosyVoiceTimeoutError(f"音色克隆任务超时（{timeout}秒）: task_id={task_id}")
-
-            response = self._call_api(
-                method="GET",
-                path=f"/tasks/{task_id}",
-                timeout=30.0,
-            )
-
-            output = response.get("output", {})
-            status = output.get("task_status", "").upper()
-
-            if status == "SUCCEEDED":
-                voice_id = output.get("voice_id", "")
-                if not voice_id:
-                    raise CosyVoiceError(f"音色克隆任务成功但未返回 voice_id: {response}")
-                return {"voice_id": voice_id}
-            elif status == "FAILED":
-                error_msg = output.get("message", "未知错误")
-                raise CosyVoiceError(f"音色克隆任务失败: {error_msg}")
-            elif status in ("PENDING", "RUNNING"):
-                # 继续轮询
-                time.sleep(self.POLL_INTERVAL)
-                attempts += 1
-            else:
-                raise CosyVoiceError(f"未知的任务状态: {status}")
-
-        raise CosyVoiceTimeoutError(f"音色克隆任务轮询次数超限: task_id={task_id}")
+        # 否则轮询
+        result = self.poll_clone_task(voice_id, timeout=timeout)
+        return CloneResult(voice_id=result["voice_id"], request_id=request_id)
 
     # ── 语音合成 ─────────────────────────────────────────
 
@@ -388,11 +404,12 @@ class CosyVoiceService:
         sample_rate: int = 0,
         format: str = "",
         speed: float = 1.0,
+        volume: int = 50,
     ) -> dict:
-        """提交语音合成任务（非阻塞）。
+        """提交语音合成任务（同步非流式，直接返回结果）.
 
-        只提交任务到 CosyVoice API，不轮询结果。
-        返回的 dict 包含 task_id（异步）或 audio_url（同步）。
+        CosyVoice SpeechSynthesizer 非流式接口是同步的，
+        调用后直接返回音频 URL. 此方法保持与旧接口兼容.
 
         Args:
             text: 要合成的文本
@@ -400,10 +417,11 @@ class CosyVoiceService:
             sample_rate: 采样率（Hz），0 表示使用配置默认值
             format: 输出格式（mp3/wav/pcm），空表示使用配置默认值
             speed: 语速（0.5-2.0），1.0 为正常速度
+            volume: 音量（0-100），默认 50
 
         Returns:
-            dict: {"task_id": str, "audio_url": str, "request_id": str}
-                  task_id 和 audio_url 至少有一个非空
+            dict: {"audio_url": str, "request_id": str,
+                   "duration": float, "file_size": int}
 
         Raises:
             CosyVoiceError: API 调用失败
@@ -423,55 +441,54 @@ class CosyVoiceService:
             "model": self._model,
             "input": {
                 "text": text,
-            },
-            "parameters": {
                 "voice": voice_id,
-                "sample_rate": sample_rate or settings.cosyvoice_sample_rate,
                 "format": format or settings.cosyvoice_format,
+                "sample_rate": sample_rate or settings.cosyvoice_sample_rate,
                 "rate": speed,
+                "volume": volume,
             },
         }
 
         response = self._call_api(
             method="POST",
-            path="/services/aigc/text2audio/generation",
+            path="/services/audio/tts/SpeechSynthesizer",
             json=payload,
-            timeout=60.0,
+            timeout=120.0,
         )
 
         output = response.get("output", {})
-        task_id = output.get("task_id", "")
-        audio_url = output.get("audio_url", "")
+        audio = output.get("audio", {})
+        audio_url = audio.get("url", "")
         request_id = response.get("request_id", "")
 
-        if not task_id and not audio_url:
-            raise CosyVoiceError(f"CosyVoice API 未返回 task_id 或 audio_url: {response}")
+        if not audio_url:
+            raise CosyVoiceError(
+                f"CosyVoice API 未返回 audio_url: {response}"
+            )
 
         return {
-            "task_id": task_id,
+            "task_id": "",  # 同步接口无 task_id，兼容旧接口
             "audio_url": audio_url,
-            "duration": output.get("duration", 0.0),
-            "file_size": output.get("file_size", 0),
+            "duration": 0.0,  # 同步接口不返回 duration
+            "file_size": 0,  # 同步接口不返回 file_size
             "request_id": request_id,
         }
 
-    def poll_synthesize_task(self, task_id: str, timeout: float = 120.0) -> dict:
-        """轮询语音合成任务状态（公开方法）。
+    def poll_synthesize_task(
+        self, task_id: str, timeout: float = 120.0
+    ) -> dict:
+        """轮询合成任务（同步接口无需轮询，保留兼容）.
 
-        供 Celery 后台任务调用，轮询直到完成或超时。
-
-        Args:
-            task_id: CosyVoice 任务 ID
-            timeout: 超时时间（秒），默认 120
-
-        Returns:
-            dict: {"audio_url": str, "duration": float, "file_size": int}
+        CosyVoice SpeechSynthesizer 非流式接口是同步的，
+        此方法仅为保持接口兼容，实际调用时 task_id 应该为空.
 
         Raises:
-            CosyVoiceError: 任务失败
-            CosyVoiceTimeoutError: 超时
+            CosyVoiceError: 同步接口无需轮询
         """
-        return self._poll_synthesize_task(task_id, timeout=timeout)
+        raise CosyVoiceError(
+            "CosyVoice 非流式合成接口是同步的，无需轮询. "
+            "请直接使用 submit_synthesize_task()."
+        )
 
     def synthesize_speech(
         self,
@@ -480,11 +497,13 @@ class CosyVoiceService:
         sample_rate: int = 0,
         format: str = "",
         speed: float = 1.0,
+        volume: int = 50,
         timeout: float = 120.0,
     ) -> SynthesizeResult:
-        """语音合成。
+        """语音合成（同步非流式）.
 
-        提交语音合成任务到 CosyVoice API，并轮询直到完成或超时。
+        调用百炼 CosyVoice SpeechSynthesizer 非流式接口，
+        直接返回合成音频 URL.
 
         Args:
             text: 要合成的文本
@@ -492,129 +511,53 @@ class CosyVoiceService:
             sample_rate: 采样率（Hz），0 表示使用配置默认值
             format: 输出格式（mp3/wav/pcm），空表示使用配置默认值
             speed: 语速（0.5-2.0），1.0 为正常速度
-            timeout: 超时时间（秒）
+            volume: 音量（0-100），默认 50
+            timeout: 超时时间（秒），保留参数兼容
 
         Returns:
             SynthesizeResult: 合成结果，包含 audio_url
 
         Raises:
             CosyVoiceError: API 调用失败
-            CosyVoiceTimeoutError: 超时
             CosyVoiceAuthError: 认证失败
             ValueError: 参数无效
         """
-        if not text:
-            raise ValueError("text 不能为空")
-        if not voice_id:
-            raise ValueError("voice_id 不能为空")
-        if not self._api_key:
-            raise CosyVoiceAuthError("CosyVoice API Key 未配置")
-
-        settings = get_shared_settings()
-
-        # 构建请求
-        payload = {
-            "model": self._model,
-            "input": {
-                "text": text,
-            },
-            "parameters": {
-                "voice": voice_id,
-                "sample_rate": sample_rate or settings.cosyvoice_sample_rate,
-                "format": format or settings.cosyvoice_format,
-                "rate": speed,
-            },
-        }
-
-        # 调用 API
-        response = self._call_api(
-            method="POST",
-            path="/services/aigc/text2audio/generation",
-            json=payload,
-            timeout=timeout,
+        result = self.submit_synthesize_task(
+            text=text,
+            voice_id=voice_id,
+            sample_rate=sample_rate,
+            format=format,
+            speed=speed,
+            volume=volume,
         )
 
-        # 解析响应
-        output = response.get("output", {})
-
-        # 检查是否有 task_id（异步模式）
-        task_id = output.get("task_id")
-        audio_url = output.get("audio_url")
-
-        if task_id:
-            # 异步模式：轮询任务状态
-            result = self._poll_synthesize_task(task_id, timeout)
-            return SynthesizeResult(
-                audio_url=result["audio_url"],
-                duration=result.get("duration", 0.0),
-                file_size=result.get("file_size", 0),
-                request_id=response.get("request_id", ""),
-            )
-        elif audio_url:
-            # 同步模式：直接返回结果
-            return SynthesizeResult(
-                audio_url=audio_url,
-                duration=output.get("duration", 0.0),
-                file_size=output.get("file_size", 0),
-                request_id=response.get("request_id", ""),
-            )
-        else:
-            raise CosyVoiceError(f"CosyVoice API 未返回 audio_url 或 task_id: {response}")
-
-    def _poll_synthesize_task(self, task_id: str, timeout: float) -> dict:
-        """轮询语音合成任务状态。
-
-        Args:
-            task_id: 任务 ID
-            timeout: 超时时间（秒）
-
-        Returns:
-            任务结果字典
-
-        Raises:
-            CosyVoiceError: 任务失败
-            CosyVoiceTimeoutError: 超时
-        """
-        start_time = time.time()
-        attempts = 0
-
-        while attempts < self.MAX_POLL_ATTEMPTS:
-            elapsed = time.time() - start_time
-            if elapsed > timeout:
-                raise CosyVoiceTimeoutError(f"语音合成任务超时（{timeout}秒）: task_id={task_id}")
-
-            response = self._call_api(
-                method="GET",
-                path=f"/tasks/{task_id}",
-                timeout=30.0,
-            )
-
-            output = response.get("output", {})
-            status = output.get("task_status", "").upper()
-
-            if status == "SUCCEEDED":
-                audio_url = output.get("audio_url", "")
-                if not audio_url:
-                    raise CosyVoiceError(f"语音合成任务成功但未返回 audio_url: {response}")
-                return {
-                    "audio_url": audio_url,
-                    "duration": output.get("duration", 0.0),
-                    "file_size": output.get("file_size", 0),
-                }
-            elif status == "FAILED":
-                error_msg = output.get("message", "未知错误")
-                raise CosyVoiceError(f"语音合成任务失败: {error_msg}")
-            elif status in ("PENDING", "RUNNING"):
-                # 继续轮询
-                time.sleep(self.POLL_INTERVAL)
-                attempts += 1
-            else:
-                raise CosyVoiceError(f"未知的任务状态: {status}")
-
-        raise CosyVoiceTimeoutError(f"语音合成任务轮询次数超限: task_id={task_id}")
+        return SynthesizeResult(
+            audio_url=result["audio_url"],
+            duration=result.get("duration", 0.0),
+            file_size=result.get("file_size", 0),
+            request_id=result.get("request_id", ""),
+        )
 
     # ── 内部方法 ─────────────────────────────────────────
 
+    def _sanitize_prefix(self, name: str) -> str:
+        """清洗音色名称为合法的 prefix（字母数字，最多10字符）.
+
+        Args:
+            name: 原始音色名称
+
+        Returns:
+            清洗后的 prefix
+        """
+        # 只保留字母和数字
+        cleaned = "".join(c for c in name if c.isalnum())
+        # 最多10字符
+        cleaned = cleaned[:10]
+        # 如果清洗后为空，用默认值
+        if not cleaned:
+            cleaned = "clone"
+        return cleaned
+
     def _call_api(
         self,
         method: str,
@@ -622,13 +565,13 @@ class CosyVoiceService:
         json: Optional[dict] = None,
         timeout: float = 30.0,
     ) -> dict:
-        """调用 CosyVoice API。
+        """调用 DashScope API.
 
-        支持重试和错误处理。
+        支持重试和错误处理.
 
         Args:
             method: HTTP 方法（GET/POST）
-            path: API 路径
+            path: API 路径（以 / 开头）
             json: 请求体
             timeout: 超时时间（秒）
 
@@ -662,25 +605,57 @@ class CosyVoiceService:
                 if response.status_code == 200:
                     return response.json()
                 elif response.status_code in (401, 403):
-                    raise CosyVoiceAuthError(f"CosyVoice API 认证失败: HTTP {response.status_code}")
+                    raise CosyVoiceAuthError(
+                        f"CosyVoice API 认证失败: HTTP {response.status_code}"
+                    )
+                elif response.status_code == 400:
+                    # 客户端错误，不重试
+                    body_text = response.text
+                    try:
+                        body = response.json()
+                        code = body.get("code", "")
+                        message = body.get("message", "")
+                        raise CosyVoiceError(
+                            f"CosyVoice API 参数错误: HTTP 400, "
+                            f"code={code}, message={message}"
+                        )
+                    except ValueError:
+                        raise CosyVoiceError(
+                            f"CosyVoice API 调用失败: HTTP 400, body={body_text}"
+                        )
                 elif response.status_code >= 500:
                     # 服务端错误，可重试
-                    last_error = CosyVoiceError(f"CosyVoice API 服务端错误: HTTP {response.status_code}")
+                    last_error = CosyVoiceError(
+                        f"CosyVoice API 服务端错误: HTTP {response.status_code}"
+                    )
                     logger.warning(
-                        f"CosyVoice API 失败 (尝试 {attempt + 1}/{self.MAX_RETRIES}): " f"HTTP {response.status_code}"
+                        "CosyVoice API 失败 (尝试 %d/%d): HTTP %d",
+                        attempt + 1,
+                        self.MAX_RETRIES,
+                        response.status_code,
                     )
                 else:
-                    # 客户端错误，不重试
+                    # 其他客户端错误，不重试
                     raise CosyVoiceError(
-                        f"CosyVoice API 调用失败: HTTP {response.status_code}, " f"body={response.text}"
+                        f"CosyVoice API 调用失败: HTTP {response.status_code}, "
+                        f"body={response.text}"
                     )
 
             except httpx.TimeoutException as e:
                 last_error = CosyVoiceTimeoutError(f"请求超时: {e}")
-                logger.warning(f"CosyVoice API 超时 (尝试 {attempt + 1}/{self.MAX_RETRIES})")
+                logger.warning(
+                    "CosyVoice API 超时 (尝试 %d/%d)",
+                    attempt + 1,
+                    self.MAX_RETRIES,
+                )
             except httpx.RequestError as e:
                 last_error = CosyVoiceError(f"请求错误: {e}")
-                logger.warning(f"CosyVoice API 请求错误 (尝试 {attempt + 1}/{self.MAX_RETRIES}): {e}")
+                logger.warning(
+                    "CosyVoice API 请求错误 (尝试 %d/%d): %s",
+                    attempt + 1,
+                    self.MAX_RETRIES,
+                    e,
+                )
 
             # 指数退避
             if attempt < self.MAX_RETRIES - 1:

packages/application/tts_job/workflow.py

@@ -14,7 +14,6 @@ import logging
 import os
 import shutil
 import tempfile
-import time
 from concurrent.futures import ThreadPoolExecutor, as_completed
 from typing import Optional
 
@@ -190,10 +189,13 @@ class TTSWorkflowService:
         return job
 
     def poll_and_process_synthesis(self, job_id: str, timeout: float = 120.0) -> TTSJob:
-        """轮询 CosyVoice 合成任务并处理结果。
+        """轮询/检查 CosyVoice 合成任务并处理结果.
 
-        从 job.metadata 获取 task_id，调用 CosyVoiceService.poll_synthesize_task()
-        轮询状态，然后通过 process_synthesis_result / process_synthesis_failure 更新 job。
+        新 CosyVoice SpeechSynthesizer 非流式接口是同步的，
+        start_synthesis 阶段通常已经完成. 此方法用于：
+        1. job 已 completed → 直接返回（同步路径已处理）
+        2. job 仍在 processing → 重新提交合成（兜底）
+        3. 分段任务 → 检查分段状态
 
         供 Celery 后台任务调用。
         """
@@ -201,22 +203,42 @@ class TTSWorkflowService:
         if job is None:
             raise TTSJobNotFoundError(f"TTS job {job_id} not found")
 
+        # 已完成直接返回（同步路径在 start_synthesis 里已处理）
+        if job.status == TTSJobStatus.COMPLETED.value:
+            logger.info(f"TTS 任务已完成，跳过轮询: job_id={job_id}")
+            return job
+
         # 检查是否为分段合成任务
         segment_task_ids = (job.metadata or {}).get("segment_task_ids", [])
         if segment_task_ids:
             return self._poll_segment_tasks(job)
 
+        # 单段模式：同步接口下通常不会走到这里，
+        # 但如果因为异常导致仍在 processing，重新提交一次
         task_id = (job.metadata or {}).get("cosyvoice_task_id", "")
-        if not task_id:
-            raise ValueError(f"TTSJob {job_id} has no cosyvoice_task_id in metadata")
 
-        result = self.cosyvoice_service.poll_synthesize_task(task_id, timeout=timeout)
-        return self.process_synthesis_result(
-            job_id,
-            audio_url=result["audio_url"],
-            duration=result.get("duration", 0.0),
-            file_size=result.get("file_size", 0),
-        )
+        # 新接口（同步）：没有 task_id，重新合成
+        if not task_id:
+            logger.info(
+                f"TTS 任务无 task_id，重新同步合成: job_id={job_id}"
+            )
+            return self._resynthesize_and_complete(job)
+
+        # 旧接口遗留的 task_id，尝试轮询（兼容过渡）
+        try:
+            result = self.cosyvoice_service.poll_synthesize_task(task_id, timeout=timeout)
+            return self.process_synthesis_result(
+                job_id,
+                audio_url=result["audio_url"],
+                duration=result.get("duration", 0.0),
+                file_size=result.get("file_size", 0),
+            )
+        except CosyVoiceError:
+            # 旧接口轮询失败，重新同步合成
+            logger.warning(
+                f"旧 task_id 轮询失败，重新同步合成: job_id={job_id}, task_id={task_id}"
+            )
+            return self._resynthesize_and_complete(job)
 
     def process_synthesis_result(
         self,
@@ -257,6 +279,40 @@ class TTSWorkflowService:
         logger.info(f"TTS 合成成功: job_id={job_id}, audio_url={permanent_url}")
         return job
 
+    def _resynthesize_and_complete(self, job: TTSJob) -> TTSJob:
+        """重新同步合成并完成任务（兜底路径）.
+
+        当 poll_and_process_synthesis 发现 job 仍在 processing 且无 task_id 时，
+        重新调用同步合成接口，转存 OSS 后标记完成。
+        """
+        try:
+            # 从 metadata 读取合成参数（兼容旧数据，无则用默认值）
+            job_metadata = job.metadata or {}
+            speed = float(job_metadata.get("speed", 1.0))
+            volume = int(job_metadata.get("volume", 50))
+
+            result = self.cosyvoice_service.submit_synthesize_task(
+                text=job.input_text,
+                voice_id=job.voice_id,
+                sample_rate=job.sample_rate,
+                format=job.format,
+                speed=speed,
+                volume=volume,
+            )
+            audio_url = result.get("audio_url", "")
+            if not audio_url:
+                raise CosyVoiceError("重新合成未返回 audio_url")
+
+            return self.process_synthesis_result(
+                job.id,
+                audio_url=audio_url,
+                duration=result.get("duration", 0.0),
+                file_size=result.get("file_size", 0),
+            )
+        except Exception as e:
+            logger.error(f"重新同步合成失败: job_id={job.id}, error={e}")
+            return self.process_synthesis_failure(job.id, str(e))
+
     def process_synthesis_failure(self, job_id: str, error_message: str) -> TTSJob:
         """处理合成失败结果。
 
@@ -429,69 +485,108 @@ class TTSWorkflowService:
             shutil.rmtree(temp_dir, ignore_errors=True)
 
     def _poll_segment_tasks(self, job: TTSJob) -> TTSJob:
-        """轮询所有分段异步任务，全部完成后合并音频。"""
+        """分段任务完成检查（适配新同步接口）.
+
+        新 CosyVoice SpeechSynthesizer 非流式接口为同步接口，
+        分段任务在提交时应已同步返回 audio_url。
+        若历史任务处于 processing 且有 segment_task_ids 但缺少 audio_url，
+        则对缺失分段重新同步合成，全部完成后合并音频。
+        """
         segment_task_ids: list[str] = (job.metadata or {}).get("segment_task_ids", [])
         segment_audio_urls: list[str] = (job.metadata or {}).get("segment_audio_urls", [])
         segment_count = len(segment_task_ids)
 
-        poll_start = time.monotonic()
-        poll_timeout = 300.0  # 分段任务超时更长
-        poll_interval = 2.0
+        if segment_count == 0:
+            logger.warning(f"分段任务无 task_id: job_id={job.id}")
+            self._handle_segment_failure(job, "分段任务数据异常：无分段信息")
+            return self.repository.get(job.id)
 
-        while time.monotonic() - poll_start < poll_timeout:
-            all_done = True
-            results: list[dict | None] = [None] * segment_count
+        # 从 metadata 读取合成参数
+        job_metadata = job.metadata or {}
+        speed = float(job_metadata.get("speed", 1.0))
+        volume = int(job_metadata.get("volume", 50))
 
-            for idx, task_id in enumerate(segment_task_ids):
-                # 已经有音频的分段跳过轮询
-                if idx < len(segment_audio_urls) and segment_audio_urls[idx]:
-                    results[idx] = {
-                        "audio_url": segment_audio_urls[idx],
-                        "duration": 0.0,
-                        "file_size": 0,
-                    }
-                    continue
+        # 分段文本（用于缺失段重新合成）
+        segments = split_text(job.input_text, max_chars=_SEGMENT_THRESHOLD)
 
-                try:
-                    result = self.cosyvoice_service.poll_synthesize_task(task_id, timeout=poll_timeout)
-                    results[idx] = result
-                except Exception as e:
-                    logger.error(f"分段任务轮询失败: job_id={job.id}, " f"segment={idx}, error={e}")
-                    self._handle_segment_failure(job, f"分段 {idx + 1} 轮询失败: {e}")
-                    return self.repository.get(job.id)
+        results: list[dict | None] = [None] * segment_count
 
-                if results[idx] is None:
-                    all_done = False
+        # 已有音频的分段直接用
+        for idx in range(segment_count):
+            if idx < len(segment_audio_urls) and segment_audio_urls[idx]:
+                results[idx] = {
+                    "audio_url": segment_audio_urls[idx],
+                    "duration": 0.0,
+                    "file_size": 0,
+                }
 
-            if all_done and all(r is not None for r in results):
-                # 所有分段完成，下载合并
-                try:
-                    merged_data, total_duration = self._download_and_merge_segments(results, job)
+        # 找出缺失音频的分段索引
+        missing_indices = [i for i in range(segment_count) if results[i] is None]
 
-                    # 转存 OSS
-                    permanent_url, storage_key = self._upload_merged_to_oss(
-                        merged_data, job.user_id, job.id, job.format
+        if missing_indices:
+            logger.info(
+                f"分段任务重新合成缺失段: job_id={job.id}, "
+                f"缺失={len(missing_indices)}/{segment_count}"
+            )
+            # 并发重新合成缺失分段
+            max_workers = min(len(missing_indices), _MAX_SEGMENT_WORKERS)
+            with ThreadPoolExecutor(max_workers=max_workers) as executor:
+                future_to_idx = {}
+                for idx in missing_indices:
+                    segment_text = segments[idx] if idx < len(segments) else ""
+                    future = executor.submit(
+                        self.cosyvoice_service.submit_synthesize_task,
+                        text=segment_text,
+                        voice_id=job.voice_id,
+                        sample_rate=job.sample_rate,
+                        format=job.format,
+                        speed=speed,
+                        volume=volume,
                     )
+                    future_to_idx[future] = idx
 
-                    job.mark_completed(
-                        output_audio_url=permanent_url,
-                        output_audio_key=storage_key,
-                        duration=total_duration,
-                        file_size=len(merged_data),
-                    )
-                    job = self.repository.update(job)
-                    logger.info(f"分段合成轮询完成: job_id={job.id}, " f"merged_size={len(merged_data)}")
-                    return job
+                for future in as_completed(future_to_idx):
+                    idx = future_to_idx[future]
+                    try:
+                        results[idx] = future.result()
+                    except Exception as e:
+                        logger.error(
+                            f"分段重新合成失败: job_id={job.id}, "
+                            f"segment={idx}, error={e}"
+                        )
+                        self._handle_segment_failure(
+                            job, f"分段 {idx + 1} 重新合成失败: {e}"
+                        )
+                        return self.repository.get(job.id)
 
-                except Exception as e:
-                    self._handle_segment_failure(job, f"分段合并失败: {e}")
-                    return self.repository.get(job.id)
+        # 所有分段完成，下载合并
+        if all(r is not None for r in results):
+            try:
+                merged_data, total_duration = self._download_and_merge_segments(results, job)
 
-            # 等待后重试
-            time.sleep(poll_interval)
+                permanent_url, storage_key = self._upload_merged_to_oss(
+                    merged_data, job.user_id, job.id, job.format
+                )
 
-        # 超时
-        self._handle_segment_failure(job, "分段合成轮询超时（300 秒）")
+                job.mark_completed(
+                    output_audio_url=permanent_url,
+                    output_audio_key=storage_key,
+                    duration=total_duration,
+                    file_size=len(merged_data),
+                )
+                job = self.repository.update(job)
+                logger.info(
+                    f"分段合成完成（重新合成路径）: job_id={job.id}, "
+                    f"merged_size={len(merged_data)}"
+                )
+                return job
+
+            except Exception as e:
+                self._handle_segment_failure(job, f"分段合并失败: {e}")
+                return self.repository.get(job.id)
+
+        # 理论上不会到这里（全部重新合成要么成功要么失败）
+        self._handle_segment_failure(job, "分段合成结果不完整")
         return self.repository.get(job.id)
 
     def _handle_segment_failure(self, job: TTSJob, error_message: str) -> None:

packages/application/voice_clone/workflow.py

@@ -115,14 +115,16 @@ class VoiceCloneWorkflowService:
                     language=language,
                 )
 
-                # 4. 保存 task_id / voice_id 到 metadata
+                # 4. 保存 voice_id / request_id 到 metadata
+                # 注意：key 保留 cosyvoice_task_id 以兼容旧数据，实际存的是 voice_id
                 task_metadata = dict(profile.metadata)
-                task_metadata["cosyvoice_task_id"] = submit_result.get("task_id", "")
+                task_metadata["cosyvoice_task_id"] = submit_result.get("voice_id", "")
                 task_metadata["cosyvoice_request_id"] = submit_result.get("request_id", "")
 
-                # 如果 CosyVoice 同步返回了 voice_id，直接标记 ready
+                # 如果 CosyVoice 直接返回了 OK 状态，直接标记 ready
                 voice_id = submit_result.get("voice_id", "")
-                if voice_id:
+                status = submit_result.get("status", "").upper()
+                if voice_id and status == "OK":
                     profile.mark_ready(voice_id)
                     profile.metadata = task_metadata
                     profile = self.repository.update(profile)
@@ -131,7 +133,7 @@ class VoiceCloneWorkflowService:
 
                 profile.metadata = task_metadata
                 profile = self.repository.update(profile)
-                logger.info(f"音色克隆任务已提交: profile_id={profile.id}, " f"task_id={submit_result.get('task_id')}")
+                logger.info(f"音色克隆任务已提交: profile_id={profile.id}, " f"voice_id={submit_result.get('voice_id')}")
 
             except (CosyVoiceError, CosyVoiceAuthError) as e:
                 # CosyVoice 提交失败，标记为 failed
@@ -248,11 +250,12 @@ class VoiceCloneWorkflowService:
                 )
 
                 task_metadata = dict(profile.metadata)
-                task_metadata["cosyvoice_task_id"] = submit_result.get("task_id", "")
+                task_metadata["cosyvoice_task_id"] = submit_result.get("voice_id", "")
                 task_metadata["cosyvoice_request_id"] = submit_result.get("request_id", "")
 
                 voice_id = submit_result.get("voice_id", "")
-                if voice_id:
+                status = submit_result.get("status", "").upper()
+                if voice_id and status == "OK":
                     profile.mark_ready(voice_id)
                     profile.metadata = task_metadata
                     profile = self.repository.update(profile)

packages/domain/preset_voices.py

@@ -16,7 +16,7 @@ class PresetVoice:
     """预置音色定义。
 
     Attributes:
-        voice_id: CosyVoice 模型音色名（如 longxiaochun）
+        voice_id: CosyVoice 模型音色名（如 longxiaochun_v3）
         name: 中文展示名
         description: 音色描述
         gender: 性别（male/female）
@@ -49,7 +49,7 @@ class PresetVoice:
 # 预置音色列表（阿里云 CosyVoice 真实可用音色）
 PRESET_VOICES: list[PresetVoice] = [
     PresetVoice(
-        voice_id="longxiaochun",
+        voice_id="longxiaochun_v3",
         name="龙小淳",
         description="温柔女声，适合情感类内容",
         gender="female",
@@ -57,7 +57,7 @@ PRESET_VOICES: list[PresetVoice] = [
         tags=["温柔", "女声", "情感"],
     ),
     PresetVoice(
-        voice_id="longxiaoxia",
+        voice_id="longxiaoxia_v3",
         name="龙小夏",
         description="知性女声，适合新闻播报",
         gender="female",
@@ -65,7 +65,7 @@ PRESET_VOICES: list[PresetVoice] = [
         tags=["知性", "女声", "播报"],
     ),
     PresetVoice(
-        voice_id="longxiaochen",
+        voice_id="longxiaochen_v3",
         name="龙小晨",
         description="磁性男声，适合有声书",
         gender="male",
@@ -73,7 +73,7 @@ PRESET_VOICES: list[PresetVoice] = [
         tags=["磁性", "男声", "有声书"],
     ),
     PresetVoice(
-        voice_id="longyue",
+        voice_id="longyue_v3",
         name="龙悦",
         description="甜美女声，适合广告配音",
         gender="female",
@@ -81,7 +81,7 @@ PRESET_VOICES: list[PresetVoice] = [
         tags=["甜美", "女声", "广告"],
     ),
     PresetVoice(
-        voice_id="longshu",
+        voice_id="longshu_v3",
         name="龙书",
         description="沉稳男声，适合教育讲解",
         gender="male",
@@ -89,7 +89,7 @@ PRESET_VOICES: list[PresetVoice] = [
         tags=["沉稳", "男声", "教育"],
     ),
     PresetVoice(
-        voice_id="longjing",
+        voice_id="longjing_v3",
         name="龙静",
         description="优雅女声，适合纪录片解说",
         gender="female",
@@ -97,7 +97,7 @@ PRESET_VOICES: list[PresetVoice] = [
         tags=["优雅", "女声", "纪录片"],
     ),
     PresetVoice(
-        voice_id="longbo",
+        voice_id="longbo_v3",
         name="龙博",
         description="浑厚男声，适合科技类内容",
         gender="male",
@@ -105,7 +105,7 @@ PRESET_VOICES: list[PresetVoice] = [
         tags=["浑厚", "男声", "科技"],
     ),
     PresetVoice(
-        voice_id="longtian",
+        voice_id="longtian_v3",
         name="龙甜",
         description="活泼女声，适合短视频配音",
         gender="female",

packages/shared/config.py

@@ -29,13 +29,15 @@ class SharedSettings(BaseSettings):
     oss_access_key_secret: str = ""
     oss_bucket_name: str = "xiaoxia-autocut"
 
-    # CosyVoice (阿里云语音合成)
+    # CosyVoice (阿里云百炼语音合成)
     cosyvoice_api_key: str = ""
-    cosyvoice_base_url: str = "https://dashscope.aliyuncs.com/api/v1/services/aigc/text2audio"
-    cosyvoice_model: str = "cosyvoice-v1"
-    cosyvoice_voice: str = "longxiaochun"  # 默认音色
+    cosyvoice_base_url: str = "https://dashscope.aliyuncs.com/api/v1"
+    cosyvoice_model: str = "cosyvoice-v3.5-plus"
+    cosyvoice_voice: str = "longxiaochun_v3"  # 默认音色
     cosyvoice_sample_rate: int = 22050
     cosyvoice_format: str = "mp3"  # 输出格式：mp3/wav/pcm
+    # 音色克隆模型名（固定为 voice-enrollment）
+    cosyvoice_clone_model: str = "voice-enrollment"
 
     # Environment
     environment: str = "development"

tests/unit/test_preset_voices.py

@@ -64,7 +64,7 @@ class TestPresetVoice:
     def test_preset_voice_to_dict(self) -> None:
         """序列化。"""
         voice = PresetVoice(
-            voice_id="longxiaochun",
+            voice_id="longxiaochun_v3",
             name="龙小淳",
             description="温柔女声",
             gender="female",
@@ -73,7 +73,7 @@ class TestPresetVoice:
 
         result = voice.to_dict()
 
-        assert result["voice_id"] == "longxiaochun"
+        assert result["voice_id"] == "longxiaochun_v3"
         assert result["name"] == "龙小淳"
         assert result["description"] == "温柔女声"
         assert result["gender"] == "female"
@@ -127,14 +127,14 @@ class TestPresetVoicesConfig:
     def test_cosyvoice_voice_ids(self) -> None:
         """音色 ID 应为 CosyVoice 真实可用的音色名。"""
         expected_ids = {
-            "longxiaochun",
-            "longxiaoxia",
-            "longxiaochen",
-            "longyue",
-            "longshu",
-            "longjing",
-            "longbo",
-            "longtian",
+            "longxiaochun_v3",
+            "longxiaoxia_v3",
+            "longxiaochen_v3",
+            "longyue_v3",
+            "longshu_v3",
+            "longjing_v3",
+            "longbo_v3",
+            "longtian_v3",
         }
         actual_ids = {v.voice_id for v in PRESET_VOICES}
         assert actual_ids == expected_ids
@@ -164,10 +164,10 @@ class TestPresetVoiceHelpers:
 
     def test_get_preset_voice_by_id_found(self) -> None:
         """按 ID 查找存在的音色。"""
-        voice = get_preset_voice_by_id("longxiaochun")
+        voice = get_preset_voice_by_id("longxiaochun_v3")
         assert voice is not None
         assert voice.name == "龙小淳"
-        assert voice.voice_id == "longxiaochun"
+        assert voice.voice_id == "longxiaochun_v3"
 
     def test_get_preset_voice_by_id_not_found(self) -> None:
         """按 ID 查找不存在的音色。"""
@@ -176,9 +176,9 @@ class TestPresetVoiceHelpers:
 
     def test_is_preset_voice_true(self) -> None:
         """判断预置音色返回 True。"""
-        assert is_preset_voice("longxiaochun") is True
-        assert is_preset_voice("longxiaoxia") is True
-        assert is_preset_voice("longbo") is True
+        assert is_preset_voice("longxiaochun_v3") is True
+        assert is_preset_voice("longxiaoxia_v3") is True
+        assert is_preset_voice("longbo_v3") is True
 
     def test_is_preset_voice_false(self) -> None:
         """判断非预置音色返回 False。"""

tests/unit/test_tts_job.py

@@ -15,7 +15,7 @@ class TestTTSJobCreate:
         job = TTSJob.create(
             user_id="user_001",
             input_text="这是一段测试文本",
-            voice_id="longxiaochun",
+            voice_id="longxiaochun_v3",
             voice_model="cosyvoice-v1",
             project_id="project_001",
             voice_clone_profile_id="profile_001",
@@ -26,7 +26,7 @@ class TestTTSJobCreate:
         assert job.id
         assert job.user_id == "user_001"
         assert job.input_text == "这是一段测试文本"
-        assert job.voice_id == "longxiaochun"
+        assert job.voice_id == "longxiaochun_v3"
         assert job.voice_model == "cosyvoice-v1"
         assert job.project_id == "project_001"
         assert job.voice_clone_profile_id == "profile_001"
@@ -291,7 +291,7 @@ class TestTTSJobToDict:
         job = TTSJob.create(
             user_id="user_001",
             input_text="测试文本",
-            voice_id="longxiaochun",
+            voice_id="longxiaochun_v3",
             voice_model="cosyvoice-v1",
             project_id="project_001",
             voice_clone_profile_id="profile_001",
@@ -306,7 +306,7 @@ class TestTTSJobToDict:
         assert result["id"] == job.id
         assert result["user_id"] == "user_001"
         assert result["input_text"] == "测试文本"
-        assert result["voice_id"] == "longxiaochun"
+        assert result["voice_id"] == "longxiaochun_v3"
         assert result["voice_model"] == "cosyvoice-v1"
         assert result["project_id"] == "project_001"
         assert result["voice_clone_profile_id"] == "profile_001"

tests/unit/test_tts_segment_synthesis.py

@@ -353,16 +353,11 @@ class TestHandleSegmentFailure:
 
 
 class TestPollSegmentTasks:
-    """测试 _poll_segment_tasks 异步轮询。"""
+    """测试 _poll_segment_tasks 分段缺失重新合成（适配同步接口）。"""
 
-    @patch("packages.application.tts_job.workflow.time")
     @patch("packages.application.tts_job.workflow.httpx")
-    def test_all_segments_done(self, mock_httpx: MagicMock, mock_time: MagicMock) -> None:
-        """所有分段完成后合并并标记完成。"""
-        # Mock time.monotonic 让循环只执行一次
-        mock_time.monotonic.side_effect = [0.0, 1.0, 2.0]
-        mock_time.sleep = MagicMock()
-
+    def test_all_segments_done(self, mock_httpx: MagicMock) -> None:
+        """所有分段缺少 audio_url 时重新同步合成，合并后标记完成。"""
         # Mock 下载分段音频
         mock_resp = MagicMock()
         mock_resp.content = b"seg audio"
@@ -370,7 +365,7 @@ class TestPollSegmentTasks:
         mock_httpx.get.return_value = mock_resp
 
         service = MagicMock(spec=CosyVoiceService)
-        service.poll_synthesize_task.side_effect = [
+        service.submit_synthesize_task.side_effect = [
             {"audio_url": "https://temp.com/seg1.mp3", "duration": 2.0, "file_size": 100},
             {"audio_url": "https://temp.com/seg2.mp3", "duration": 3.0, "file_size": 200},
         ]
@@ -381,6 +376,8 @@ class TestPollSegmentTasks:
         repo = MagicMock()
         job = _make_job(
             status=TTSJobStatus.PROCESSING,
+            # 长文本触发分段，用于重新合成时切分
+            input_text="这是一段很长的测试文本。" * 30,
             metadata={
                 "segment_task_ids": ["task_1", "task_2"],
                 "segment_audio_urls": ["", ""],
@@ -400,19 +397,18 @@ class TestPollSegmentTasks:
             result = workflow._poll_segment_tasks(job)
 
         assert result.status == TTSJobStatus.COMPLETED
+        # 两个缺失分段都重新合成了
+        assert service.submit_synthesize_task.call_count == 2
 
-    @patch("packages.application.tts_job.workflow.time")
-    def test_segment_poll_failure(self, mock_time: MagicMock) -> None:
-        """分段轮询失败时标记 job failed。"""
-        mock_time.monotonic.side_effect = [0.0, 1.0]
-        mock_time.sleep = MagicMock()
-
+    def test_segment_resynthesis_failure(self) -> None:
+        """分段重新合成失败时标记 job failed。"""
         service = MagicMock(spec=CosyVoiceService)
-        service.poll_synthesize_task.side_effect = CosyVoiceError("Poll failed")
+        service.submit_synthesize_task.side_effect = CosyVoiceError("Synthesis failed")
 
         repo = MagicMock()
         job = _make_job(
             status=TTSJobStatus.PROCESSING,
+            input_text="这是一段很长的测试文本。" * 30,
             metadata={
                 "segment_task_ids": ["task_1"],
                 "segment_audio_urls": [""],
@@ -426,6 +422,51 @@ class TestPollSegmentTasks:
 
         assert result.status == TTSJobStatus.FAILED
 
+    @patch("packages.application.tts_job.workflow.httpx")
+    def test_partial_audio_urls_reuse_existing(self, mock_httpx: MagicMock) -> None:
+        """部分分段已有 audio_url 时直接复用，缺失的重新合成。"""
+        mock_resp = MagicMock()
+        mock_resp.content = b"seg audio"
+        mock_resp.raise_for_status.return_value = None
+        mock_httpx.get.return_value = mock_resp
+
+        service = MagicMock(spec=CosyVoiceService)
+        # 只有 1 个分段需要重新合成
+        service.submit_synthesize_task.return_value = {
+            "audio_url": "https://temp.com/seg2.mp3",
+            "duration": 3.0,
+            "file_size": 200,
+        }
+
+        storage = MagicMock()
+        storage.upload_file.return_value = "https://oss.example.com/merged.mp3"
+
+        repo = MagicMock()
+        job = _make_job(
+            status=TTSJobStatus.PROCESSING,
+            input_text="这是一段很长的测试文本。" * 30,
+            metadata={
+                "segment_task_ids": ["task_1", "task_2"],
+                "segment_audio_urls": ["https://temp.com/seg1.mp3", ""],
+                "segment_format": "mp3",
+            },
+        )
+        repo.get.return_value = job
+        repo.update.side_effect = lambda j: j
+
+        workflow = _make_workflow(cosyvoice_service=service, repo=repo, storage=storage)
+
+        with patch("packages.application.tts_job.workflow.AudioMerger") as MockMerger:
+            mock_merger = MagicMock()
+            mock_merger.merge.return_value = b"merged data"
+            MockMerger.return_value = mock_merger
+
+            result = workflow._poll_segment_tasks(job)
+
+        assert result.status == TTSJobStatus.COMPLETED
+        # 只有 1 个缺失分段被重新合成
+        assert service.submit_synthesize_task.call_count == 1
+
 
 class TestPollAndProcessSynthesisSegmentDetection:
     """测试 poll_and_process_synthesis 正确识别分段任务。"""

tests/unit/test_voice_clone_workflow.py

@@ -57,15 +57,15 @@ def _make_service(
 class TestStartClone:
     """测试 start_clone 方法。"""
 
-    def test_start_clone_with_async_task(self) -> None:
-        """异步模式：提交任务后返回 processing 状态的 profile。"""
+    def test_start_clone_with_deploying(self) -> None:
+        """提交克隆后返回 DEPLOYING 状态，profile 保持 processing。"""
         mock_repo = MagicMock()
         mock_cosyvoice = MagicMock(spec=CosyVoiceService)
 
-        # CosyVoice 返回 task_id（异步模式）
+        # CosyVoice 返回 voice_id + DEPLOYING 状态（需轮询）
         mock_cosyvoice.submit_clone_task.return_value = {
-            "task_id": "task-abc",
-            "voice_id": "",
+            "voice_id": "voice-abc",
+            "status": "DEPLOYING",
             "request_id": "req-123",
         }
 
@@ -81,21 +81,21 @@ class TestStartClone:
         )
 
         assert profile.status == VoiceCloneStatus.PROCESSING
-        assert profile.metadata["cosyvoice_task_id"] == "task-abc"
+        assert profile.metadata["cosyvoice_task_id"] == "voice-abc"
         assert profile.metadata["cosyvoice_request_id"] == "req-123"
         mock_cosyvoice.submit_clone_task.assert_called_once()
         assert mock_repo.create.call_count == 1
-        # update 至少调用 2 次：mark_processing + 保存 task_id
+        # update 至少调用 2 次：mark_processing + 保存 voice_id
         assert mock_repo.update.call_count >= 2
 
-    def test_start_clone_with_sync_result(self) -> None:
-        """同步模式：CosyVoice 直接返回 voice_id，profile 变为 ready。"""
+    def test_start_clone_with_ok_status(self) -> None:
+        """CosyVoice 直接返回 OK 状态，profile 变为 ready。"""
         mock_repo = MagicMock()
         mock_cosyvoice = MagicMock(spec=CosyVoiceService)
 
         mock_cosyvoice.submit_clone_task.return_value = {
-            "task_id": "",
             "voice_id": "voice-sync-123",
+            "status": "OK",
             "request_id": "req-456",
         }
 
@@ -244,8 +244,8 @@ class TestRetryClone:
         mock_repo.update.side_effect = lambda p: p
 
         mock_cosyvoice.submit_clone_task.return_value = {
-            "task_id": "task-retry",
-            "voice_id": "",
+            "voice_id": "voice-retry",
+            "status": "DEPLOYING",
             "request_id": "req-retry",
         }
 
@@ -253,11 +253,11 @@ class TestRetryClone:
         result = service.retry_clone(profile.id, "user-123")
 
         assert result.status == VoiceCloneStatus.PROCESSING
-        assert result.metadata["cosyvoice_task_id"] == "task-retry"
+        assert result.metadata["cosyvoice_task_id"] == "voice-retry"
         assert result.retry_count == 2  # prepare_retry 增加了一次
 
-    def test_retry_clone_with_sync_result(self) -> None:
-        """重试成功，同步模式。"""
+    def test_retry_clone_with_ok_status(self) -> None:
+        """重试成功，直接返回 OK 状态。"""
         mock_repo = MagicMock()
         mock_cosyvoice = MagicMock(spec=CosyVoiceService)
 
@@ -266,8 +266,8 @@ class TestRetryClone:
         mock_repo.update.side_effect = lambda p: p
 
         mock_cosyvoice.submit_clone_task.return_value = {
-            "task_id": "",
             "voice_id": "voice-retry-sync",
+            "status": "OK",
             "request_id": "req-retry",
         }