LaunchAI ModelsSeptember 17, 2026

Wispr's Canto speech model tops real-world dictation WER test

Wispr Advanced Interfaces Lab's Canto posted the lowest word error rate among models tested from Google, OpenAI, AssemblyAI, and Deepgram on 10 hours of real-world dictation from 2,300+ speakers. Train and test speakers were strictly separated to avoid overfitting.

1 source

More stories today

Open the live feed