HOT RADAR
llmHEAT 12.6°

AI CURATED HOT STORY · 11/16/2026

WisprFlow Debuts In-House Speech Model Canto 1 at 3.4% WER

1 reports archived1 independent sourcesupdated 11/16/2026, 05:59:28
Synthesis & Latest Updates
1 Sources Cross-Validated

Speech-input startup WisprFlow has hired Ariya Rastrow, a founding Alexa scientist, as chief science officer and shifted from rented speech engines to training its own models. Its first in-house model, Canto 1, was trained on noisy real-world dictation and posted a 3.4% word error rate across roughly 10 hours of real Flow dictations from over 2,300 speakers, beating every competing model Wispr tested while cutting dictations that take over a second to appear by 54%.

LATEST/Speech-input startup WisprFlow has hired Ariya Rastrow, a founding Alexa scientist, as chief science officer and shifted from rented speech engines to training its own models. Its first in-house model, Canto 1, was trained on noisy real-world dictation and posted a 3.4% word error rate across roughly 10 hours of real Flow dictations from over 2,300 speakers, beating every competing model Wispr tested while cutting dictations that take over a second to appear by 54%.

Hourly heat curve

2 fully observed hours
Fewer than 3 fully observed hours; trend not plotted yet.

Coverage timeline

Total 1 reports · Latest first
  1. Rohan Paul (@rohanpaul_ai)T2·71 pts
    • WisprFlow hired founding Alexa scientist Ariya Rastrow as CSO and moved its speech stack from rented engines to in-house models
    • Canto 1 hit a 3.4% word error rate on ~10 hours of real dictations from 2,300+ speakers, beating all competitors Wispr tested
    • The model targets noisy environments and per-user vocabulary, cutting dictations slower than one second by 54%