Fact · Accessibility and health
62percent relative word error rate improvement
The personalized models in the 2019 Google study gave a 62 percent relative reduction in word error rate on speech from people with ALS.
Checked against its source on October 16, 2026
In the source's own words
“We train personalized models that achieve 62% and 35% relative WER improvement on these two groups”
Personalizing ASR for Dysarthric and Accented Speech with Limited Data, Joel Shor et al., Interspeech 2019, 2019, Abstract.
Open the original- What it measures
- Relative improvement compares the personalized model with the starting model on the same speakers. It is not the same as the absolute error rate.
- What it does not show
- A relative figure says nothing about the absolute error rate, which depends on how impaired the speaker is. The same abstract gives the absolute rates in the fact on mild and more serious dysarthria.