Why not just run Whisper locally?
The fair question every technical buyer asks — answered with the trade-offs, not around them.
Whisper is open source, free, and runs on your own machine. Lven Instant is free too, so the fair question is which one to run.
Run the numbers on your own hardware first. Whisper's headline accuracy lives in its large model — roughly three gigabytes in memory before you say a word, and on consumer-grade hardware it transcribes slower than you speak. Step down to the sizes that fit ordinary hardware and you step down in accuracy too.
Our engine doesn't ask you to choose. On consumer-grade hardware it transcribes at a real-time factor of 0.166 — about six times faster than you speak — with a 3.76% word error rate, on the CPU. In the same measurement whisper-large-v3 needed 3.36 GB of video memory and did not reach realtime. The full method is in the research note.
And the model is the smaller half of the answer. whisper.cpp gives you a transcription engine; dictation is everything around it — capturing audio, a global hotkey, landing text at your cursor in whatever app has focus, cleaning stutters and restoring punctuation without rewriting you. Lven ships all of that, working, on Windows, Linux, and Android.
Lven is for people who just want to talk.