feat: 桌面端语音输入（右 Alt 录音 → 智谱 ASR 转写）

[wishfay]

Summary

在聊天输入框按右 Alt 开始录音,再按一次结束,语音经识别后转成文字插入输入框(不自动发送,可编辑)。语音识别模型在设置 → 供应商」配置,默认用智谱 glm-asr-2512 测试。

方案

调用链:前端录音 → 本地 server(3456)→ server 用配置的 Key 调用语音识别 provider →
返回文字 → 插入输入框。

走 server 中转:前端 CSP 禁止直连外部域名(只允许 localhost),且与现有 provider/Tavily
key 一致——外部 Key 由 server 持有并发起调用。

关键点

• 智谱云端 ASR 仅支持 wav/mp3,Chromium 的 MediaRecorder 产不出这两种格式,因此前端用
  ScriptProcessorNode 采 PCM 并内联编码成 WAV(无第三方依赖)
• document 级监听右 Alt,toggle
  录音/停止;录音/识别中/错误状态有指示器,错误透传真实原因
• 「设置 → 供应商」拆分为**「语言模型」(原有 provider,驱动对话与
  Agent)与「语音模型」**(语音转文字配置)两个分区

改动

Server:新增 server/api/voice.ts(POST /api/voice/transcribe)、路由注册、设置
schema 增加 voiceInput
桌面端:新增 api/voice.ts + hooks/useVoiceInput.ts,接入 ChatInput(光标插入 +
指示器),Settings.tsx 增加语音配置区,store/types/i18n(5 语言)同步

验证

• cd desktop && bun run lint(tsc)通过
• 服务端启动无报错,/api/voice/transcribe 路由实测 400/405/404 符合预期
• 录音→转写→插入的端到端建议在真实环境(填智谱 API Key)下进一步验证

备注

智谱 GLM-ASR 仅支持 wav/mp3(≤25MB),故前端必须录成 WAV。macOS 可能需在 src-tauri 的
Info.plist 补 NSMicrophoneUsageDescription(本次未涉及)。

[github-actions]

PR quality triage

Changed areas: area:cli-core, area:desktop, area:release, area:server

CLI core policy: Blocked by policy until a maintainer applies allow-cli-core-change and approves the PR.

Missing-test policy: Blocked by policy until a maintainer applies allow-missing-tests or matching tests are added.

Coverage baseline policy: No coverage-baseline policy block detected.

CLI core files:

• src/utils/settings/types.ts

Coverage policy files:

• none

Expected checks:

• change-policy
• desktop-checks
• server-checks
• desktop-native-checks
• coverage-checks

Test coverage signals:

• BLOCKING unless allow-missing-tests is applied: Desktop product files changed without a desktop test file in the PR.
• BLOCKING unless allow-missing-tests is applied: Server product files changed without a server test file in the PR.
• BLOCKING unless allow-missing-tests is applied: Agent/runtime product files changed without a tools/utils test file in the PR.
• Agent/model runtime path changed: use mock/request-shape tests in PR and maintainer live-model smoke before release.

Risk notes:

• Desktop state/API layer changed: verify store persistence, WebSocket behavior, and startup errors.

Hard merge gates still come from GitHub Actions, not AI review.

Dosu handoff: Dosu can be used as the AI reviewer for risk explanation, missing-test prompts, and maintainer Q&A. If it does not comment automatically from the PR template, ask:

@dosubot review this PR for changed-area risk, missing tests, docs impact, desktop startup risk, and CLI core impact.