The shift from typing to voice input is accelerating fast. Wispr Flow just hired Ariya Rastrow (ex-Amazon, scaled Alexa's speech team from 0→400 engineers) as CSO. He launched an internal research lab 4 months ago.

Two months later they shipped Canto 1, their proprietary speech model. Real-world metrics since deployment:
• Desktop users correcting 13% fewer dictations
• Slow dictation events down 54%
• Daily dictation volume up 7%
• Ranked #1 on VoiceArena's dictation benchmark this week

The technical breakthrough isn't just accuracy—it's trust. When correction rate drops below a psychological threshold, users stop editing and typing becomes the slower option. That behavioral flip is the actual product moat.

Companies building foundational speech models in-house (not relying on third-party APIs) will control the voice interface layer. Wispr is betting that owning the model stack matters more than distribution for this transition.