dsh-plugin-voice-input-demo

It is often faster to think aloud than to type when you are composing a longer Chinese reply in the DeepSeek Harness Web UI. dsh-plugin-voice-input-demo is a minimal plugin built for exactly that moment: it adds a single microphone button to the right of the input field. Tap to start recording, tap again or let the single-utterance recognition end naturally to stop, and your speech is transcribed into text and appended to the input box. Nothing is sent automatically; you can still edit, trim, or reword the text and hit send only when you are satisfied.

The plugin relies entirely on the browser-native SpeechRecognition / webkitSpeechRecognition API. It introduces no extra SDK, stores no audio, and is fixed to Simplified Chinese (zh-CN). It does not alter any other part of DeepSeek Harness and requires no server-side changes. Keep in mind that the API is unavailable in some browsers (Chrome is recommended), and certain browsers may route audio to their own speech service, so fully offline processing is not guaranteed.

If you think more naturally in spoken form and want the lowest-friction way to get Chinese text into the DeepSeek Harness web interface, this plugin offers a clean five-step flow: press, speak, release, edit, send. No configuration, no setup, just a button.