Fact · Languages and accents
19.7percent word error rate
On the EdAcc accent test set of conversations between friends, the best-performing English speech recognition model, trained on 680 thousand hours of transcribed data, averaged a word error rate of 19.7 percent, against 2.7 percent on US English clean read speech.
Checked against its source on October 17, 2026
In the source's own words
“The best performing model, trained on 680 thousand hours of transcribed data, obtains an average of 19.7% word error rate (WER) -- in contrast to the 2.7% WER obtained when evaluated on US English clean read speech.”
The Edinburgh International Accents of English Corpus: Towards the Democratization of English ASR, Ramon Sanabria et al., arXiv (accepted to IEEE ICASSP 2023), 2023, Abstract.
Open the original- What it measures
- EdAcc contains almost 40 hours of video calls between friends, with speakers from many English varieties. The authors tested the latest public and commercial English models on it, and the best one was trained on 680 thousand hours of transcribed data.
- What it does not show
- The 2.7 percent comparison is with clean read speech from US English speakers, which is easier than conversation. The 19.7 percent figure is an average over all accent groups, not a result for any one accent.