dsh-voice-chat

Typing every reply in the DSH Web GUI works, but long sessions get visually tiring and voice input lacks a smooth interaction loop. dsh-voice-chat closes that gap: tap the mic to speak, and the AI reads its reply back to you out loud, delivering an experience close to a Doubao-style voice assistant.

The plugin covers three pillars of capability. On the input side, a 2.5-second pause automatically ends recording, transcribes the speech, and sends it. On the output side, replies are read aloud via TTS; a paraphrase-and-read mode (off by default) lets the LLM condense longer answers before speaking them, staying faithful and concise. The ASR layer supports SiliconFlow (SenseVoice), Groq (Whisper), Xiaomi MiMo, and any OpenAI-compatible endpoint. The TTS layer supports Edge TTS (free, no API key required), MiMo TTS, and custom OpenAI-compatible services. Single-channel playback ensures only one voice plays at a time, a replay guard prevents re-reading on session re-entry, and Ctrl+Shift+Space toggles the microphone in one press.

Built for users who prefer voice over typing and for developers who want flexible multi-engine switching with low-latency playback. All settings take effect immediately from the DSH settings panel without a restart. Requires Node.js 22 or later and a mainstream browser; Edge TTS works out of the box with zero configuration.