Fact · Languages and accents
6.3points
A 2026 study found that scoring orthographic variants as correct, using its OIWER method, lowers error rates for Indian-language speech recognition by an average of 6.3 points.
Checked against its source on October 17, 2026
In the source's own words
“reduces pessimistic error rates (an average improvement of 6.3 points)”
Towards Orthographically-Informed Evaluation of Speech Recognition Systems for Indian Languages, Kaushal Santosh Bhogale et al., arXiv preprint (AI4Bharat, IIT Madras), 2026, Abstract.
Open the original- What it measures
- Hindi and other Indian languages allow several spellings of the same word, and code-mixed English words are often written in more than one way. Standard word error rate counts a correct word with a different spelling as wrong.
- What it does not show
- The 6.3 points is an average across the study's benchmarks. Measured error rates in your own dictation can differ, and the authors' method is a research benchmark rather than a product setting.