Speak your mind, offline.
ExtMind Voice is a 100 %-on-device voice keyboard for Android. Speak in Chinese, English, Japanese, Korean or Cantonese — and edit the result with an on-device LLM. No cloud. No account. No analytics.
What it does
🎤 Bilingual dictation
Speak naturally. Mid-sentence code-switching between Chinese and English just works — auto-language detection is built into the on-device model.
✏ Polish
Light edit pass: fixes mishearings, punctuation, capitalisation. The pen icon fills with ink as the local LLM works through your text.
✨ Rewrite
Deeper restructuring of what you just said. Same on-device LLM, longer maximum output.
🌐 Hand-off to your keyboard
When you need to type, tap to switch to your regular keyboard. Speak when you want to speak.
Why on-device
- Privacy. Your voice never leaves your phone. There is no server to leak from, no account to compromise.
- Cost. No subscription. No per-minute API charges. You own the models on disk.
- Reliability. Works on a plane, on a subway, in a tunnel, in a country with patchy mobile data.
- Latency. Voice activity detection commits the moment you stop speaking — ~150 ms, no network round-trip.
Under the hood
SenseVoice-small for final transcription, Online Paraformer for live partials, ct-transformer for punctuation, Silero VAD for endpoint detection, Qwen2.5-1.5B-Instruct (via llama.cpp) for Polish and Rewrite. All open source. All running on your phone's CPU.
Get it
ExtMind Voice is in closed alpha. Email hello@extmind.ai if you want in.
Read
- Privacy policy — explicit, plain-English
- Source on GitHub (placeholder)