Fact · Languages and accents
5.65 to 43.03percentage points
For the open source models in the test, the gap between the error rate on Svarah and on native English read speech from the Speech Accent Archive ranged from 5.65 to 43.03 percentage points.
Checked against its source on October 17, 2026
In the source's own words
“the gap between the WERs on Svarah and SAA L1 ranges from 5.65 - 43.03”
Svarah: Evaluating English ASR Systems on Indian Accents, Tahir Javed et al., arXiv, 2023, Section 5, Results and Discussions.
Open the original- What it measures
- The authors use the gap to show that the claims of human-level accuracy in English recognition do not hold for Indian speakers.
- What it does not show
- The gap is between two different test sets, so it is a measure of the difference, not a single cause. The range is across models, from smallest to largest gap.