What is on-device transcription?
On-device transcription is speech-to-text that runs entirely on your own hardware: a speech model on your CPU or GPU converts microphone audio into text locally, so the audio is never uploaded to a server. After the model is set up, transcription itself can run without a network connection.
How it differs from cloud speech-to-text
Cloud dictation streams your microphone audio to a server, which returns text. That can offer strong accuracy on weak hardware, but every word you speak transits and is processed off your machine, and nothing works when the network doesn't.
On-device flips the trade: the model runs where you are. Privacy becomes structural — audio has nowhere to go — and dictation keeps working on a plane, on hostile Wi-Fi, or in a security-conscious workplace. The cost is that your hardware does the work, so there is a realistic minimum spec.
What 'offline' means here
In any credible on-device product, the transcription path is local. Setup still downloads the model once, and licensed products periodically validate accounts or entitlements over the network. The distinction that matters: none of those requests carry your audio or your transcript text.
Where it's heading
Speech models keep shrinking while consumer chips keep gaining AI throughput, so the accuracy gap between local and cloud keeps narrowing. Dictation is following the same arc storage and photography followed: from a service back into the device.