Fact · Languages and accents
The Svarah authors say the medium and large versions of Whisper gave single-digit word error rates on Indian English, and note that their size makes them hard to deploy in India.
Checked against its source on October 17, 2026
In the source's own words
“the medium and large variants of Whisper give single-digit WERs”
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 add that, given the large size of these models, it may be difficult to deploy them in the Indian context.
- What it does not show
- The single-digit result is for the 2023 test set. The smaller Whisper models in the same table scored higher error rates, so the model size you can run on your device changes the result.