Preprint
Towards Orthographically-Informed Evaluation of Speech Recognition Systems for Indian Languages
Kaushal Santosh Bhogale et al., arXiv preprint (AI4Bharat, IIT Madras), 2026
Read the originalRead on October 17, 2026
- What it is
- Argues that standard word error rate overstates recognition errors for Indian languages, because acceptable spelling variants and code-mixed words count as errors. It proposes an orthographically informed measure, OIWER, and reports that it lowers error rates by an average of 6.3 points.
- How much weight it carries
- A preprint whose authors built the measure they propose. The point that word error rate counts spelling variants as errors is a general observation; the 6.3-point figure applies to the authors benchmarks.
Facts taken from this source
- 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.
“reduces pessimistic error rates (an average improvement of 6.3 points)” (Abstract)