dsh-voice-input

Your keyboard is not the only way to type. dsh-voice-input turns voice input into a system-wide habit: press and hold a hotkey anywhere on your desktop, speak, release, and the transcribed text lands at your cursor — in a browser, a chat app, a word processor, a search box, anywhere text is accepted. No window switching, no app launching, the whole desktop becomes an input field.

Under the hood, a locally running Whisper small model handles recognition with solid Chinese accuracy, automatically using CUDA acceleration when an NVIDIA GPU is detected, delivering results in roughly one to two seconds. The model is lazily loaded, so it consumes no GPU or RAM until you actually speak. Beyond transcription, the plugin offers pure audio recording with WAV export, a searchable history of every clip and transcript, and a tray settings panel for adjusting hotkeys, toggling push-to-talk mode, selecting model size, language, inference device, and recording directory — changes take effect instantly and survive plugin updates.

A good fit for writers and knowledge workers who produce heavy text volume and want to trade keystrokes for words; for privacy-conscious users who would rather keep voice data on their own hardware than ship it to a cloud API; and for anyone who simply likes the idea of a small microphone icon in the system tray meaning: press, hold, and just talk it out.