Wispr AI Lab Releases Canto Speech Model for Real-World Dictation
September 17, 2026
Canto is a speech model optimized for real-time dictation in noisy environments like commutes or busy offices. In evaluations against Google, OpenAI, AssemblyAI, and Deepgram, Canto achieved the lowest word error rate for audio recorded via laptop microphones and earbuds in uncontrolled conditions.
HOW THIS AFFECTS YOU
●
builderYou can integrate higher-accuracy transcription into voice-first applications used in loud environments.
●
researcherThis model provides a new benchmark for robustness against non-stationary background noise in speech recognition.