Vendor page
Personalized ASR Models from a Large and Diverse Disordered Speech Dataset
Google Research Blog, 2021
Read the originalRead on October 16, 2026
- What it is
- Google Research's September 2021 blog post summarizing two Interspeech 2021 papers from Project Euphonia: the growth of its disordered speech corpus and personalized speech recognition models trained on it.
- How much weight it carries
- Written by the researchers who built the dataset; a primary source for the corpus size. A company blog, not peer-reviewed itself, though it summarizes peer-reviewed papers.
Facts taken from this source
- Google trained personalized recognizers for about 430 speakers with speech impairments, each of whom had recorded at least 300 utterances.
“We trained personalized ASR models for ~430 speakers who recorded at least 300 utterances” (Results)
- By August 2021, Google's Project Euphonia corpus of disordered speech held over 1 million utterances, more than 1,400 hours, from 1,330 speakers.
“This effort has grown Euphonia's corpus to over 1 million utterances, comprising over 1400 hours from 1330 speakers (as of August 2021).” (Project Euphonia data collection)
- 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.
“the median WER for short phrases from the home automation domain dropped from around 89% to 13%” (Results)