Fact · Accessibility and health
13percent median word error rate
Google reports that personalized models cut the median word error rate on severely impaired speech from about 89 percent to 13 percent for short home-automation phrases.
Checked against its source on October 16, 2026
In the source's own words
“the median WER for short phrases from the home automation domain dropped from around 89% to 13%”
Personalized ASR Models from a Large and Diverse Disordered Speech Dataset, Google Research Blog, 2021, Results.
Open the original- What it measures
- The result comes from Project Euphonia's personalized models for about 430 speakers, tested on short home-automation phrases such as 'Turn on the TV.'
- What it does not show
- The phrases were short and scripted, so it does not predict the error rate for long dictation. The blog post is Google's summary of its paper.