We built our own speech model. It beat Google and OpenAI on real dictation (Sponsor)
Most voice apps rent someone else's speech model. Wispr Flow built its own, called Canto, trained for how people actually talk: mid-thought, half-whispered, with Slack pinging in the background.
Tested on 10 hours of real Wispr Flow dictations from more than 2,300 speakers, Canto had the lowest word error rate of every model tested, including Google, OpenAI, AssemblyAI and Deepgram.
For marketers, that means briefs, campaign emails and prompts come out right the first time, including client and brand names from your dictionary. Talk through the draft and get clean, formatted text anywhere you type.
Mac, Windows, iPhone and Android.
