dsh-voice-funasr

In an era where voice input boosts efficiency, privacy and offline reliability remain top concerns. Traditional cloud-based services require uploading audio to remote servers, which not only fails without a stable internet connection but also raises data security issues. dsh-voice-funasr tackles this by localizing the entire speech recognition process, enabling smooth voice input even in offline or privacy-sensitive scenarios.

The plugin's core strength lies in its local-first architecture. It integrates an optimized FunASR engine (using paraformer-large, FSMN-VAD, and ct-punc models), performing all inference on-device for highly accurate Chinese transcription. After transcription, an optional two-stage polishing powered by LLM automatically fixes filler words and slips of the tongue, producing cleaner text. If the local engine becomes unavailable, it seamlessly falls back to the browser's built-in Web Speech API, ensuring uninterrupted input.

dsh-voice-funasr is ideal for DSH Web UI users who prioritize privacy, professionals often working in offline or unstable network environments, and anyone demanding high accuracy for Chinese voice recognition. Audio data flows only in memory and is never stored permanently, safeguarding your conversation privacy from the ground up. Whether for daily chats or focused creation, it serves as a reliable and secure voice input companion.