QT-Chen/dsh-mic-input
DSH Web ?????????:??? Web Speech API ????,????/??????????????????Microphone voic
Microphone voice input for the DSH Web UI, entirely in the browser: adds a mic button to the composer tool row with a red pulsing ring while listening, live interim transcription into the input, duplicate-text-free draft rebuilding, auto-continue on silence (reconnects up to 6 silent ends so long sentences are not cut off), smart punctuation completion (？/！/。 by language, or preserve/remove modes), language selection (auto/zh-CN/en-US), optional auto-send when dictation ends, and a settings row under 设置 → 通用 → 语音输入. Audio never leaves the browser except to the browser's own speech service.
Install
dsh plugin --profile web add github:QT-Chen/dsh-mic-inputGitHub install per README (English edition): dsh plugin --profile web add github:QT-Chen/dsh-mic-input, restart dsh web (or refresh the page), allow microphone permission on 127.0.0.1/localhost. README also documents dsh plugin --profile web add dsh-mic-input from npm but the npm package returned 404 at 2026-08-26 verification — the GitHub path is the reliable one.
Compatibility
DeepSeek Harness web profile. Pure client plugin using the built-in Web Speech API — Microsoft Edge → Azure speech, Google Chrome → Google/Chrome speech. Zero server components, zero API keys, zero model downloads.
Details
- Repo: QT-Chen/dsh-mic-input
- Category: Web UI Enhancements
- Stars: 3
- Version: GitHub source per README (npm dsh-mic-input 404 at verify)
- Last push: 2026-08-16
- First seen: 2026-08-14
Recent updates
mic button; live transcription; silence auto-continue; smart punctuation; language selection; auto-send; bilingual error hints.
FAQ
- Does my audio leave the machine?
- No server or API keys are involved — audio goes only to the browser's own speech service (Azure for Edge, Google for Chrome).
- Why does recognition produce garbled text?
- Wrong language matching is the #1 cause — set the language explicitly (跟随系统 / 中文 / English) in the settings row.
- How do I stop recording?
- Click the mic button again; the transcript stays in the input and is auto-sent only if you enabled auto-send.
Alternatives
Anionex/dsh-suggested-replies · forrestahha/dsh-voice-input · loudMore/dsh-drop-to-path