dsh-voice-webspeech

DSH Web GUI has lacked a lightweight, ready-to-use voice input option. The popular dsh-voice-funasr plugin requires a local Python runtime and roughly a 520 MB FunASR model download, which is heavy for users who simply want to dictate in their browser. dsh-voice-webspeech takes a leaner path: it relies entirely on the browser built-in Web Speech API (Edge routes to Azure Speech, Chrome to Google Speech), with zero server-side components, zero API keys, zero model downloads, and zero Python.

The core interaction revolves around a microphone button next to the composer: hold to start listening, release to stop, with consecutive phrases accumulating seamlessly. While listening, a floating bubble shows real-time interim results above the button (toggleable). Transcribed text is appended to existing composer content by default so you can edit before sending, or switched to hands-free auto-send in settings. It supports Mandarin, Cantonese, Traditional Chinese, English, Japanese, Korean, French, German, Spanish, Russian, and more, with preferences stored locally in the browser.

This plugin is well suited for Edge or Chrome users who want fast voice typing without maintaining a local ASR engine or large model files. Audio goes directly to the browser speech service, bypassing any DSH server, never written to disk, with no telemetry. Note that Firefox and Safari are not supported because they lack the Web Speech API.