Canto: a speech model built for the real world
Summary
Wispr AI Lab announces Canto, a real-world dictation speech model that achieves the lowest word error rate among tested models on real-world Wispr Flow dictations. The article outlines evaluation results, training methodology (supervised fine-tuning followed by reinforcement learning with GRPO), contextual vocabulary usage, and plans for future work including diarization and multilingual support.