Canto: A speech model built for the real world

Wispr Advanced Interfaces Lab developed Canto, a speech model optimized for real-time dictation in noisy, real-world environments. Canto achieved the lowest word error rate among real-time transcription models when tested against competitors like Google, OpenAI, and Deepgram. It performs particularly well on low-volume speech and short dictations compared to larger, non-real-time models.