dsh-kitt-voice

Most voice plugins for the harness only handle Chinese and English, and the moment the browser window loses focus—say you have a fullscreen game or an IDE in front of it—the whole thing goes silent. dsh-kitt-voice turns voice into a real conversation: press a key, talk, the Silero endpoint detector figures out when you have stopped, the transcript is sent to the agent, the reply is read back sentence by sentence while the agent is still writing it, and the microphone picks up again for your next turn. No button between turns. It separates a human voice from a TV, an engine, or a whistle using a purpose-built detector rather than a loudness threshold, so a noisy room does not feed nonsense to your agent. A floating companion bar rides above whatever application is on screen, global hotkeys let you talk from a game or an editor, and you can interrupt a reply mid-sentence by simply talking over it.

The interface ships in Spanish, English, and Simplified Chinese independently of the language you dictate in. For speech output you can choose from 104 neural voices spanning Spanish, English, and Chinese, fall back to locally synthesised Piper voices for a fully offline setup, or use the system voice with zero configuration. Microphone and speaker are picked separately by name, reading speed runs from half to double, and a guidance vocabulary keeps English technical terms like setup or understeer intact inside Spanish sentences.

It is built for developers who want to escape the typing-only workflow, Spanish speakers who have been underserved by existing voice plugins, anyone who needs hands-free agent interaction from any frontmost application, and privacy-minded users who prefer keeping every byte of text on their own machine.