dsh-voice-chat

The DeepSeek Harness Web UI is built around text interaction, which can feel tiring over long sessions of typing questions and reading replies word by word. The dsh-voice-chat plugin weaves voice input and text-to-speech reading directly into the conversation flow, so you can ask with your voice and listen to answers, dramatically lowering the interaction cost.

The plugin runs entirely in the browser using the Web Speech API, requiring no server, no API key, and no model download. Core features include a microphone button that transcribes speech into the input box in real time (editable before sending), a speaker button next to every AI reply for on-demand TTS playback, and an auto-read mode that only announces genuinely new replies (deduplicated by message sequence number, so history and page reloads never trigger re-reading). A settings panel lets you pick a voice, adjust speech rate and pitch, and choose the recognition language from Mandarin, Cantonese, Taiwanese Mandarin, or English. Before speaking, Markdown markup is stripped automatically so the TTS never reads literal asterisks or hash symbols.

Ideal for users who prefer voice-driven interaction, spend long hours in the Web UI, or need multi-language voice recognition. All preferences persist in localStorage across page reloads, making it a zero-configuration install-and-use experience.