VoiceInput
Type with your voice. Anywhere.
Open-source dictation app for macOS and iOS. Tap once to start, tap again to stop — text pastes wherever your cursor is. Apple Speech runs locally; optional self-hosted STT and LLM polish via your own infrastructure. Built by NexAgent.
What is VoiceInput?
VoiceInput is a fast, privacy-respecting dictation utility that turns your voice into text in any app — Notes, Messages, terminals, IDEs, browsers, even native dialogs. On macOS, hold Fn+Ctrl, speak, release: a streaming partial transcript appears next to the menu-bar mic; the final text pastes into whatever you were typing in. On iOS, install the VoiceInput keyboard extension, tap the mic button, and the iFlytek/Gboard-style container-app bounce captures and transcribes your audio, then pastes back into the host field. Apple's on-device speech recognizer powers the default flow — fully offline, free, no API key. For production-grade accuracy you can swap in a self-hosted Qwen3-ASR or OpenAI-compatible Whisper endpoint. An optional LLM refiner cleans up homophone errors and (in Polish mode) restructures long dictation into paragraphs and bullet lists.
VoiceInput vs. Stock Dictation
Built for engineers, writers, and clinicians who type all day and want voice without compromise.
- Limitation
Trigger
Press F5 or globe key once, then speak — gates UI
- Limitation
STT Engine
Apple-only, no swap option
- Limitation
Privacy
Closed-source pipeline; varies by vendor
- Limitation
LLM Polish
None — raw recognition output
- Advantage
Trigger
Hold Fn+Ctrl on macOS / tap once on iOS — invisible to host app
- Advantage
STT Engine
Apple Speech default; pluggable Cloud STT (Qwen3-ASR / Whisper)
- Advantage
Privacy
Open-source, MIT licensed; STT can run 100% on your hardware
- Advantage
LLM Polish
Optional self-hosted LLM polish (paragraph + bullet structuring)
Why Engineers Love It
VoiceInput Questions
Is VoiceInput free?
Yes. The macOS app and iOS keyboard are both open source under MIT license. You can self-host the STT and LLM endpoints if you want, or use Apple's free on-device speech recognition. There's no subscription, no telemetry, no account required.
Does it work without internet?
Yes. The default macOS configuration uses Apple's on-device SFSpeechRecognizer — fully offline, supports Chinese, English, Japanese, Korean and more. Cloud STT (for higher accuracy or multilingual scenarios) is optional and points at any OpenAI-Whisper-compatible endpoint of your choice.