Merge branch 'main' into py_fix_voiceprint

2026-07-22 07:03:53 +08:00 · 2025-09-08 18:13:51 +08:00
parent 8aed23f21a 01838f598c
commit b12c1efc90
7 changed files with 184 additions and 11 deletions
@@ -0,0 +1,24 @@
+-- VOSK ASR模型供应器
+delete from `ai_model_provider` where id = 'SYSTEM_ASR_VoskASR';
+INSERT INTO `ai_model_provider` (`id`, `model_type`, `provider_code`, `name`, `fields`, `sort`, `creator`, `create_date`, `updater`, `update_date`) VALUES
+('SYSTEM_ASR_VoskASR', 'ASR', 'vosk', 'VOSK离线语音识别', '[{"key": "model_path", "type": "string", "label": "模型路径"}, {"key": "output_dir", "type": "string", "label": "输出目录"}]', 11, 1, NOW(), 1, NOW());
+
+-- VOSK ASR模型配置
+delete from `ai_model_config` where id = 'ASR_VoskASR';
+INSERT INTO `ai_model_config` VALUES ('ASR_VoskASR', 'ASR', 'VoskASR', 'VOSK离线语音识别', 0, 1, '{\"type\": \"vosk\", \"model_path\": \"\", \"output_dir\": \"tmp/\"}', NULL, NULL, 11, NULL, NULL, NULL, NULL);
+
+-- 更新VOSK ASR配置说明
+UPDATE `ai_model_config` SET 
+`doc_link` = 'https://alphacephei.com/vosk/',
+`remark` = 'VOSK ASR配置说明：
+1. VOSK是一个离线语音识别库，支持多种语言
+2. 需要先下载模型文件：https://alphacephei.com/vosk/models
+3. 中文模型推荐使用vosk-model-small-cn-0.22或vosk-model-cn-0.22
+4. 完全离线运行，无需网络连接
+5. 输出文件保存在tmp/目录
+使用步骤：
+1. 访问 https://alphacephei.com/vosk/models 下载中文模型
+2. 解压模型文件到项目目录下的models/vosk/文件夹
+3. 在配置中指定正确的模型路径
+4. 注意：VOSK中文模型输出不带标点符号，词与词之间会有空格
+' WHERE `id` = 'ASR_VoskASR';
@@ -310,3 +310,10 @@ databaseChangeLog:
        - sqlFile:
            encoding: utf8
            path: classpath:db/changelog/202509081140.sql
+  - changeSet:
+      id: 202508271113
+      author: cgd
+      changes:
+        - sqlFile:
+            encoding: utf8
+            path: classpath:db/changelog/202508271113.sql
@@ -114,7 +114,10 @@ plugins:
  # 想稳定一点就自行申请替换，每天有1000次免费调用
  # 申请地址：https://console.qweather.com/#/apps/create-key/over
  # 申请后通过这个链接可以找到自己的apihost：https://console.qweather.com/setting?lang=zh
-  get_weather: {"api_host":"mj7p3y7naa.re.qweatherapi.com", "api_key": "a861d0d5e7bf4ee1a83d9a9e4f96d4da", "default_location": "广州" }
+  get_weather:
+    api_host: "mj7p3y7naa.re.qweatherapi.com"
+    api_key: "a861d0d5e7bf4ee1a83d9a9e4f96d4da"
+    default_location: "广州"
  # 获取新闻插件的配置，这里根据需要的新闻类型传入对应的url链接，默认支持社会、科技、财经新闻
  # 更多类型的新闻列表查看 https://www.chinanews.com.cn/rss/
  get_news_from_chinanews:
@@ -392,6 +395,22 @@ ASR:
    base_url: https://api.groq.com/openai/v1/audio/transcriptions
    model_name: whisper-large-v3-turbo
    output_dir: tmp/
+  VoskASR:
+    # 官方网站：https://alphacephei.com/vosk/
+    # 配置说明：
+    # 1. VOSK是一个离线语音识别库，支持多种语言
+    # 2. 需要先下载模型文件：https://alphacephei.com/vosk/models
+    # 3. 中文模型推荐使用vosk-model-small-cn-0.22或vosk-model-cn-0.22
+    # 4. 完全离线运行，无需网络连接
+    # 5. 输出文件保存在tmp/目录
+    # 使用步骤：
+    # 1. 访问 https://alphacephei.com/vosk/models 下载对应的模型
+    # 2. 解压模型文件到项目目录下的models/vosk/文件夹
+    # 3. 在配置中指定正确的模型路径
+    # 4. 注意：VOSK中文模型输出不带标点符号，词与词之间会有空格
+    type: vosk
+    model_path: 你的模型路径，如：models/vosk/vosk-model-small-cn-0.22
+    output_dir: tmp/


  
@@ -913,4 +932,4 @@ TTS:
    audio_format: "pcm"
    # 默认音色，如需其他音色可到项目assets文件夹下注册
    voice: "jay_klee"
-    output_dir: tmp/
+    output_dir: tmp/
@@ -432,10 +432,14 @@ class ConnectionHandler:
        try:
            voiceprint_config = self.config.get("voiceprint", {})
            if voiceprint_config:
-                self.voiceprint_provider = VoiceprintProvider(voiceprint_config)
-                self.logger.bind(tag=TAG).info("声纹识别功能已在连接时动态启用")
+                voiceprint_provider = VoiceprintProvider(voiceprint_config)
+                if voiceprint_provider is not None and voiceprint_provider.enabled:
+                    self.voiceprint_provider = voiceprint_provider
+                    self.logger.bind(tag=TAG).info("声纹识别功能已在连接时动态启用")
+                else:
+                    self.logger.bind(tag=TAG).warning("声纹识别功能启用但配置不完整")
            else:
-                self.logger.bind(tag=TAG).info("声纹识别功能未启用或配置不完整")
+                self.logger.bind(tag=TAG).info("声纹识别功能未启用")
        except Exception as e:
            self.logger.bind(tag=TAG).warning(f"声纹识别初始化失败: {str(e)}")

@@ -0,0 +1,114 @@
+import os
+import json
+import time
+from typing import Optional, Tuple, List
+from .base import ASRProviderBase
+from config.logger import setup_logging
+from core.providers.asr.dto.dto import InterfaceType
+import vosk
+
+TAG = __name__
+logger = setup_logging()
+
+class ASRProvider(ASRProviderBase):
+    def __init__(self, config: dict, delete_audio_file: bool = True):
+        super().__init__()
+        self.interface_type = InterfaceType.LOCAL
+        self.model_path = config.get("model_path")
+        self.output_dir = config.get("output_dir", "tmp/")
+        self.delete_audio_file = delete_audio_file
+        
+        # 初始化VOSK模型
+        self.model = None
+        self.recognizer = None
+        self._load_model()
+        
+        # 确保输出目录存在
+        os.makedirs(self.output_dir, exist_ok=True)
+
+    def _load_model(self):
+        """加载VOSK模型"""
+        try:
+            if not os.path.exists(self.model_path):
+                raise FileNotFoundError(f"VOSK模型路径不存在: {self.model_path}")
+                
+            logger.bind(tag=TAG).info(f"正在加载VOSK模型: {self.model_path}")
+            self.model = vosk.Model(self.model_path)
+
+            # 初始化VOSK识别器（采样率必须为16kHz）
+            self.recognizer = vosk.KaldiRecognizer(self.model, 16000)
+
+            logger.bind(tag=TAG).info("VOSK模型加载成功")
+        except Exception as e:
+            logger.bind(tag=TAG).error(f"加载VOSK模型失败: {e}")
+            raise
+
+    async def speech_to_text(
+        self, audio_data: List[bytes], session_id: str, audio_format: str = "opus"
+    ) -> Tuple[Optional[str], Optional[str]]:
+        """将语音数据转换为文本"""
+        file_path = None
+        try:
+            # 检查模型是否加载成功
+            if not self.model:
+                logger.bind(tag=TAG).error("VOSK模型未加载，无法进行识别")
+                return "", None
+
+            # 解码音频（如果原始格式是Opus）
+            if audio_format == "pcm":
+                pcm_data = audio_data
+            else:
+                pcm_data = self.decode_opus(audio_data)
+                
+            if not pcm_data:
+                logger.bind(tag=TAG).warning("解码后的PCM数据为空，无法进行识别")
+                return "", None
+                
+            # 合并PCM数据
+            combined_pcm_data = b"".join(pcm_data)
+            if len(combined_pcm_data) == 0:
+                logger.bind(tag=TAG).warning("合并后的PCM数据为空")
+                return "", None
+
+            # 判断是否保存为WAV文件
+            if not self.delete_audio_file:
+                file_path = self.save_audio_to_file(pcm_data, session_id)
+
+            start_time = time.time()
+            
+            
+            # 进行识别（VOSK推荐每次送入2000字节的数据）
+            chunk_size = 2000
+            text_result = ""
+            
+            for i in range(0, len(combined_pcm_data), chunk_size):
+                chunk = combined_pcm_data[i:i+chunk_size]
+                if self.recognizer.AcceptWaveform(chunk):
+                    result = json.loads(self.recognizer.Result())
+                    text = result.get('text', '')
+                    if text:
+                        text_result += text + " "
+            
+            # 获取最终结果
+            final_result = json.loads(self.recognizer.FinalResult())
+            final_text = final_result.get('text', '')
+            if final_text:
+                text_result += final_text
+            
+            logger.bind(tag=TAG).debug(
+                f"VOSK语音识别耗时: {time.time() - start_time:.3f}s | 结果: {text_result.strip()}"
+            )
+            
+            return text_result.strip(), file_path
+            
+        except Exception as e:
+            logger.bind(tag=TAG).error(f"VOSK语音识别失败: {e}")
+            return "", None
+        finally:
+            # 文件清理逻辑
+            if self.delete_audio_file and file_path and os.path.exists(file_path):
+                try:
+                    os.remove(file_path)
+                    logger.bind(tag=TAG).debug(f"已删除临时音频文件: {file_path}")
+                except Exception as e:
+                    logger.bind(tag=TAG).error(f"文件删除失败: {file_path} | 错误: {e}")
@@ -167,14 +167,13 @@ class ServerMCPClient:

                # 建立SSEClient
                elif "url" in self.config:
+                    headers = dict(self.config.get("headers", {}))
+                    # TODO 兼容旧版本
                    if "API_ACCESS_TOKEN" in self.config:
-                        headers = {
-                            "Authorization": f"Bearer {self.config['API_ACCESS_TOKEN']}"
-                        }
-                    else:
-                        headers = {}
+                        headers["Authorization"] = f"Bearer {self.config['API_ACCESS_TOKEN']}"
+                        self.logger.bind(tag=TAG).warning(f"你正在使用旧过时的配置 API_ACCESS_TOKEN ，请在.mcp_server_settings.json中将API_ACCESS_TOKEN直接设置在headers中，例如 'Authorization': 'Bearer API_ACCESS_TOKEN'")
                    sse_r, sse_w = await stack.enter_async_context(
-                        sse_client(self.config["url"], headers=headers)
+                        sse_client(self.config["url"], headers=headers, timeout=self.config.get("timeout", 5), sse_read_timeout=self.config.get("sse_read_timeout", 60 * 5))
                    )
                    read_stream, write_stream = sse_r, sse_w

@@ -36,6 +36,12 @@
      "command": "npx",
      "args": ["-y", "@simonb97/server-win-cli"],
      "link": "https://github.com/SimonB97/win-cli-mcp-server"
+    },
+    "sse-mcp-server": {
+      "url": "http://localhost:8080/sse",
+      "headers": {
+        "Authorization": "Bearer YOUR TOKEN"
+      }
    }
  }
 }