feat: refresh model sync and MiMo speech metadata (#91)

schemas/model-spec.schema.json

@@ -18,6 +18,107 @@
   },
   "additionalProperties": false,
   "definitions": {
+    "modelOrigin": {
+      "type": "object",
+      "required": ["vendor", "canonicalModelId"],
+      "properties": {
+        "vendor": {
+          "type": "string",
+          "minLength": 1,
+          "pattern": "^[a-z0-9]+(?:-[a-z0-9]+)*$",
+          "description": "规范化原厂商 ID，如 xiaomi、openai、alibaba、minimax；不得填写当前算力 Provider ID。"
+        },
+        "canonicalModelId": {
+          "type": "string",
+          "minLength": 1,
+          "description": "原厂模型的稳定识别 ID，用于适配器匹配、诊断和日志；不替代实际路由使用的 modelName/apiModelId。"
+        },
+        "family": {
+          "type": "string",
+          "minLength": 1,
+          "description": "跨小版本稳定的模型家族，如 mimo-asr；同一协议下的后续模型版本可复用该家族语义。"
+        }
+      },
+      "additionalProperties": false
+    },
+    "speechVoice": {
+      "type": "object",
+      "required": ["id", "label"],
+      "properties": {
+        "id": {
+          "type": "string",
+          "minLength": 1,
+          "description": "传给上游语音接口的稳定音色 ID。"
+        },
+        "label": {
+          "type": "string",
+          "minLength": 1,
+          "description": "面向用户展示的音色名称。"
+        }
+      },
+      "additionalProperties": false
+    },
+    "speechModelProfile": {
+      "type": "object",
+      "required": ["protocol", "protocolVersion", "modes"],
+      "properties": {
+        "protocol": {
+          "type": "string",
+          "minLength": 1,
+          "pattern": "^[a-z0-9]+(?:-[a-z0-9]+)*$",
+          "description": "稳定语音协议族，如 xiaomi-mimo-chat-asr；模型常规迭代不得改变此值。"
+        },
+        "protocolVersion": {
+          "type": "string",
+          "pattern": "^[1-9][0-9]*$",
+          "description": "协议不兼容版本号；只有请求、认证、响应或流式状态机发生不兼容变化时才递增。"
+        },
+        "modes": {
+          "type": "array",
+          "minItems": 1,
+          "uniqueItems": true,
+          "items": { "type": "string", "enum": ["batch", "streaming"] },
+          "description": "模型协议支持的调用模式：batch（整段）和/或 streaming（流式）。"
+        },
+        "inputFormats": {
+          "type": "array",
+          "minItems": 1,
+          "uniqueItems": true,
+          "items": { "type": "string", "minLength": 1 },
+          "description": "ASR 上游接受的音频格式，如 wav、mp3、webm；格式名使用适配器可识别的规范值。"
+        },
+        "outputFormats": {
+          "type": "array",
+          "minItems": 1,
+          "uniqueItems": true,
+          "items": { "type": "string", "minLength": 1 },
+          "description": "TTS 上游可返回的音频格式，如 pcm16le、wav、mp3；运行时需与适配器和播放器能力求交集。"
+        },
+        "sampleRates": {
+          "type": "array",
+          "minItems": 1,
+          "uniqueItems": true,
+          "items": { "type": "integer", "minimum": 1 },
+          "description": "支持的采样率（Hz）；运行时不得假设固定采样率，需完成能力协商。"
+        },
+        "defaultVoice": {
+          "type": "string",
+          "minLength": 1,
+          "description": "TTS 推荐默认音色 ID；声明 voices 时必须能在 voices 中找到同 ID 条目。"
+        },
+        "voices": {
+          "type": "array",
+          "minItems": 1,
+          "items": { "$ref": "#/definitions/speechVoice" },
+          "description": "TTS 可选音色列表，统一为稳定 id 与用户可读 label。"
+        },
+        "languageHints": {
+          "type": "boolean",
+          "description": "ASR 协议是否接受语言提示参数；不是模型支持语言范围清单。"
+        }
+      },
+      "additionalProperties": false
+    },
     "modelSpec": {
       "type": "object",
       "required": ["id", "spec"],
@@ -158,9 +259,11 @@
             },
             "extra": {
               "type": "object",
-              "description": "模型内在附加配置（如 apiModelId 全名、推理预算等）。不含价格，也不得声明 Provider/接入面的 reasoning effort 能力。",
+              "description": "模型内在开放扩展：modelOrigin 描述真实原厂身份，speech 描述可跨小版本复用的语音协议；不含价格，也不得声明 Provider/接入面的 reasoning effort 能力。",
               "not": { "required": ["reasoning"] },
               "properties": {
+                "modelOrigin": { "$ref": "#/definitions/modelOrigin" },
+                "speech": { "$ref": "#/definitions/speechModelProfile" },
                 "thinkingOnly": {
                   "type": "boolean",
                   "description": "模型强制思考、无法关闭（如 GLM-5.3 系列 thinking.type 仅接受 enabled）。true 时客户端把 off 从可选档位移除，并把内部 suppressThinking 降级为 auto，避免发出必然被上游拒绝的关闭参数。只描述模型固有属性，不声明接入面支持哪些深度档位——后者只能写在 provider model 的 extra.reasoning 中。"