zemanzhang809/dsh-stt-plugin
Adds a microphone toggle to the composer tool row that dictates into the draft with the browser's Web Speech API, plus a Voice Input settings page for language, continuous dictation and auto-send.
Recognition runs entirely in the browser, so no audio ever reaches the host or the model provider — only the recognized text enters the conversation, including interim results as you speak. Preferences persist through the DSH settings service in the ui-stt namespace: recognition language, continuous dictation (phrases accumulate in the draft until you toggle off) and auto-send, which is off by default and submits as soon as a phrase finishes in single-shot mode. Slots, styles, locale dictionaries and the settings scope are all Fiber-owned and removed on unload, so disabling the plugin leaves nothing behind.
Install
dsh plugin --profile web add github:zemanzhang809/dsh-stt-pluginThe github: line from the README; there is no npm package. The npm-style line appears in the README only as the first of its documented options, marked as the registry form.
Compatibility
Needs DeepSeek Harness with profile support (run-from-source pnpm dsh web or a deployed bundle), Node.js >= 18 and pnpm >= 8 to build and install. Browser support is explicit: Chrome and Edge full, Safari partial, Firefox unsupported with the button rendered disabled.
Details
- Repo: zemanzhang809/dsh-stt-plugin
- Category: Vision & Multimodal
- Stars: 1
- Version: MIT; version not stated in the README
Alternatives
sprainJinyu/dsh-vision-link · huashenglian/dsh-omni-workstation · corrinehu/dsh-chat-imagine