Richard-Yang0130/dsh-web-speech-input
Voice input plugin for DeepSeek Harness using the browser Web Speech API
Adds voice input to the DSH Web composer: a microphone button on the right side of the input box, an animated recording pill with real elapsed time (正在听 00:04), tap again to stop. Only final recognition results are written into the ordinary draft — interim results never pollute it — and it never auto-sends; you keep the normal edit + send flow. Handles unsupported-browser, permission-denied, no-speech, and recognition-failure states; dark-theme tokens, keyboard focus, and reduced-motion preferences respected. The plugin only contributes the conversation.input.right slot control and uses the existing setDraft action — it has no way to send messages and does not touch session persistence.
Install
dsh plugin --profile web add github:Richard-Yang0130/dsh-web-speech-inputdsh plugin --profile web add github:Richard-Yang0130/dsh-web-speech-input — the package declares dsh.bundle, so installation automatically adds the speech-input row to the Web profile (no manual cordis.patch.yml edits). Then dsh web (usually http://127.0.0.1:3080). Requirements: current dsh CLI, pnpm, Node.js 22.18+. Uninstall: dsh plugin --profile web remove @local/dsh-web-speech-input. README is zh-primary with a full English section at the bottom.
Compatibility
DSH Web profile; browser Web Speech API (SpeechRecognition / webkitSpeechRecognition) — best in Chrome/Edge; Safari/Firefox may not support it. Not offline: browsers may send audio to their own recognition service. Defaults to zh-CN recognition.
Details
- Repo: Richard-Yang0130/dsh-web-speech-input
- Category: Web UI Enhancements
- Stars: 2
- Version: GitHub source (no npm package)
- Last push: 2026-08-14
- First seen: 2026-08-14
Recent updates
Web Speech API mic input; recording pill with elapsed time; final-results-only draft writes; no auto-send.
FAQ
- Which browsers work?
- Chrome/Edge support SpeechRecognition/webkitSpeechRecognition; Safari, Firefox, or enterprise-policy-restricted browsers may not.
- Does it auto-send the transcript?
- No — only final results are written to the draft; you review and send as usual.
- Is it offline?
- No — the plugin itself stores no audio and calls no paid API, but the browser may send audio to its vendor's recognition service.
Alternatives
forrestahha/dsh-voice-input · YYTbit/dsh-plugin-pi-bridge · anweat/dsh-voice-webspeech