feat: 新增腾讯云ASR服务，支持按配置切换ASR提供商

- 新增tencent_asr_service.py腾讯云一句话识别
- 优化asr_service.py
- 更新services/__init__.py按ASR_PROVIDER切换whisper/tencent
- 更新requirements.txt

Co-authored-by: Cursor <cursoragent@cursor.com>

api/requirements.txt

@@ -44,7 +44,7 @@ faster-whisper>=1.0.0
 # Image Processing
 Pillow>=10.0.0
 
-# Tencent Cloud SMS
+# Tencent Cloud SDK（全量包，包含 SMS、ASR 语音识别等模块）
 tencentcloud-sdk-python>=3.0.1000
 
 openai

api/services/__init__.py

@@ -1,7 +1,28 @@
 """
 服务模块
+通过环境变量 ASR_PROVIDER 选择 ASR 实现：
+  - ASR_PROVIDER: whisper（默认，本地 faster-whisper）| tencent（腾讯云一句话识别）
 """
-from .asr_service import asr_service
+import os
+import logging
+
+logger = logging.getLogger(__name__)
+
+# =============================================================================
+# ASR Provider 选择
+# =============================================================================
+ASR_PROVIDER = os.getenv("ASR_PROVIDER", "whisper").lower()
+
+if ASR_PROVIDER == "tencent":
+    from .tencent_asr_service import tencent_asr_service as asr_service
+    logger.info("ASR Provider: 腾讯云一句话识别")
+else:
+    from .asr_service import asr_service
+    logger.info("ASR Provider: faster-whisper（本地）")
+
+# =============================================================================
+# TTS 及其他服务
+# =============================================================================
 from .tts_service import tts_service
 from .llm_service import llm_service
 from .redis_service import redis_service
@@ -12,4 +33,3 @@ __all__ = [
     "llm_service",
     "redis_service",
 ]
-

api/services/asr_service.py

@@ -16,11 +16,16 @@ logger = logging.getLogger(__name__)
 ASR_MODEL_SIZE = os.getenv("ASR_MODEL_SIZE", "small")
 ASR_DEVICE = os.getenv("ASR_DEVICE", "auto")  # auto, cpu, cuda
 ASR_COMPUTE_TYPE = os.getenv("ASR_COMPUTE_TYPE", "auto")  # auto, int8, float16, float32
-# 镜像内预置模型目录，设置后直接使用本地模型不联网下载（与 Dockerfile 中 download_root 一致）
+# 模型缓存目录：每次启动优先从该目录加载，不设置则使用默认本地路径（api/models/whisper）
+# 设置 ASR_MODEL_CACHE_DIR 时仅使用本地模型不联网（与 Dockerfile 中 download_root 一致）
 ASR_MODEL_CACHE_DIR = os.getenv("ASR_MODEL_CACHE_DIR")
+# 默认本地缓存目录（相对 api 目录），确保每次启动都先从本地加载
+_DEFAULT_ASR_CACHE_DIR = os.path.normpath(
+    os.path.join(os.path.dirname(os.path.abspath(__file__)), "..", "models", "whisper")
+)
 
 
-class ASRService:
+class WhisperASRService:
     """
     ASR 服务（语音转文字）
     使用 faster-whisper 本地模型
@@ -60,8 +65,12 @@ class ASRService:
                 else:
                     compute_type = "int8"  # CPU 使用 int8 量化，速度更快
             
-            download_root = ASR_MODEL_CACHE_DIR if ASR_MODEL_CACHE_DIR else None
-            local_files_only = bool(ASR_MODEL_CACHE_DIR)
+            # 每次启动都先从本地目录加载：优先用环境变量，否则用默认 api/models/whisper
+            download_root = ASR_MODEL_CACHE_DIR if ASR_MODEL_CACHE_DIR else _DEFAULT_ASR_CACHE_DIR
+            local_files_only = bool(ASR_MODEL_CACHE_DIR)  # 仅当显式设置缓存目录时禁止联网（如 Docker）
+            if not os.path.isdir(download_root):
+                os.makedirs(download_root, exist_ok=True)
+            logger.info(f"Whisper 模型从本地加载: download_root={download_root}, local_files_only={local_files_only}")
             self.model = WhisperModel(
                 ASR_MODEL_SIZE,
                 device=device,
@@ -156,4 +165,4 @@ class ASRService:
 
 
 # 全局实例
-asr_service = ASRService()
+asr_service = WhisperASRService()

api/services/tencent_asr_service.py

@@ -0,0 +1,113 @@
+"""
+ASR 服务：语音转文字（腾讯云一句话识别）
+使用腾讯云 ASR API 进行语音识别
+"""
+import base64
+import logging
+import os
+import uuid
+from typing import Optional
+
+logger = logging.getLogger(__name__)
+
+
+class TencentASRService:
+    """
+    ASR 服务（语音转文字）
+    使用腾讯云一句话识别 API（SentenceRecognition）
+    文档：https://cloud.tencent.com/document/product/1093/35646
+    """
+
+    def __init__(self):
+        self._secret_id = os.getenv("TENCENT_SECRET_ID", "")
+        self._secret_key = os.getenv("TENCENT_SECRET_KEY", "")
+        self._app_id = os.getenv("TENCENT_ASR_APP_ID", "")
+        self._ready = bool(self._secret_id and self._secret_key)
+        self._client = None
+
+        if not self._ready:
+            logger.warning("腾讯云 ASR 未配置：缺少 TENCENT_SECRET_ID 或 TENCENT_SECRET_KEY")
+
+    def _get_client(self):
+        """懒加载腾讯云 ASR 客户端"""
+        if self._client is not None:
+            return self._client
+
+        try:
+            from tencentcloud.common import credential
+            from tencentcloud.common.profile.client_profile import ClientProfile
+            from tencentcloud.common.profile.http_profile import HttpProfile
+            from tencentcloud.asr.v20190614 import asr_client
+
+            cred = credential.Credential(self._secret_id, self._secret_key)
+
+            http_profile = HttpProfile()
+            http_profile.endpoint = "asr.tencentcloudapi.com"
+
+            client_profile = ClientProfile()
+            client_profile.httpProfile = http_profile
+
+            self._client = asr_client.AsrClient(cred, "", client_profile)
+            logger.info("腾讯云 ASR 客户端初始化成功")
+            return self._client
+        except Exception as e:
+            logger.error(f"腾讯云 ASR 客户端初始化失败: {e}", exc_info=True)
+            return None
+
+    def ensure_ready(self) -> bool:
+        """
+        确保 ASR 服务已就绪。
+        腾讯云 ASR 是远程 API，无需预加载模型，仅检查凭证配置。
+        """
+        if not self._ready:
+            return False
+        # 尝试初始化客户端，验证 SDK 可用
+        return self._get_client() is not None
+
+    def is_ready(self) -> bool:
+        """检查 ASR 服务是否可用。"""
+        return self._ready
+
+    async def transcribe(self, audio_base64: str) -> Optional[str]:
+        """
+        转写音频为文字（腾讯云一句话识别）
+
+        Args:
+            audio_base64: Base64 编码的音频数据
+
+        Returns:
+            转写文本，失败时返回错误信息
+        """
+        if not self._ready:
+            return "转写失败: 腾讯云 ASR 未配置"
+
+        client = self._get_client()
+        if not client:
+            return "转写失败: 腾讯云 ASR 客户端初始化失败"
+
+        try:
+            from tencentcloud.asr.v20190614 import models
+
+            req = models.SentenceRecognitionRequest()
+            req.EngSerViceType = "16k_zh"  # 16k 中文普通话
+            req.SourceType = 1  # 1 = 语音数据的 Base64 编码
+            req.VoiceFormat = "m4a"  # 音频格式
+            req.Data = audio_base64
+            req.DataLen = len(base64.b64decode(audio_base64))
+
+            resp = client.SentenceRecognition(req)
+            result = resp.Result
+
+            logger.info(
+                f"腾讯云 ASR 转写完成: 文本长度={len(result) if result else 0}"
+            )
+
+            return result.strip() if result else ""
+
+        except Exception as e:
+            logger.error(f"腾讯云 ASR 转写失败: {e}", exc_info=True)
+            return f"转写失败: {str(e)}"
+
+
+# 全局实例
+tencent_asr_service = TencentASRService()