# Azure Pronunciation Assessment 资源/key 就绪诊断与 smoke test ## Azure Pronunciation Assessment 诊断流程样例 ### 工具入口 ```bash python tools/azure_speech_pronunciation_diag.py \ --audio samples/pronunciation_smoke.wav \ --reference-text "Good morning, welcome to the lesson." ``` ### 输入来源 | 输入 | 来源 | 用途 | | --- | --- | --- | | `AZURE_SPEECH_KEY` | 环境变量或 `az cognitiveservices account keys list` | REST 鉴权 | | `AZURE_SPEECH_REGION` | 环境变量或 Speech resource location | 拼接 endpoint | | sample WAV | 本地样例音频 | short-audio smoke test | | reference text | CLI 参数 | Pronunciation Assessment header | | Azure account | `az account show` | 判断登录态、tenant、subscription | ### 核心检查步骤 1. 检查 `az` 是否存在；不存在则提示安装 Azure CLI。 2. 检查 `az account show`；失败则提示 `az login`。 3. 检查当前账号、tenant、subscription；如果出现 `not in Microsoft Services tenant`，直接判定为租户/账号权限阻塞。 4. 优先使用 `AZURE_SPEECH_KEY` 和 `AZURE_SPEECH_REGION`；缺失时尝试从指定 resource group/resource name 读取 Speech key。 5. 发送 REST short-audio 请求，带 `Pronunciation-Assessment` header。 6. 输出结构化结论：`READY`、`BLOCKED_BY_TENANT`、`MISSING_KEY_REGION`、`RESOURCE_NOT_FOUND`、`SMOKE_FAILED`。 ### 成功信号 ```text READY account: [email protected] subscription: xxx region: eastus smoke: HTTP 200 pronunciation_score: 86.0 ``` ### 失败信号 ```text BLOCKED_BY_TENANT reason: [email protected] is not in Microsoft Services tenant next_action: switch to an account with Azure subscription access, invite the account as external user, or manually provide AZURE_SPEECH_KEY/AZURE_SPEECH_REGION and rerun smoke ``` ### 建议落地 把这个流程变成 product-bu 的工作区 skill。以后用户说“Azure Pronunciation Assessment 还没通 / key 在哪 / 测一下接口”时，agent 先跑诊断入口，只有诊断明确需要交互登录或租户授权时才把动作交还给用户。