dsh-voice-control

Typing long prompts and reading replies word by word in every DSH Web session adds up to real friction. dsh-voice-control wires the browser-native Web Speech API into the DeepSeek Harness conversation loop, so you can speak a prompt, hear the reply, and treat your desktop as a lightweight voice-first workspace.

On the input side, a microphone button sits right next to the send key; push-to-talk recording avoids picking up TTS output and forming an echo loop, and recognised text is either dispatched immediately or dropped into the draft for a quick edit before sending. On the output side, the latest assistant reply is cleaned of Markdown and read aloud once the stream settles, with voices matched to your browser and OS in real time (Edge users additionally get access to Microsoft natural online voices). Speed, voice, and auto-send preferences are all adjustable from a right-click panel and persisted in localStorage.

Everything runs entirely in the browser with zero external dependencies, no backend, and no API keys, so you are ready to talk the moment the page loads. It is a great fit for Chrome and Edge users who want a voice-centric workflow with DeepSeek Harness without spinning up extra services. Firefox is not supported yet because it lacks a Web Speech recognition implementation.