Fact · Languages and accents
7.2percent word error rate
On Svarah, a 2023 benchmark of English from Indian speakers, the large version of OpenAI's Whisper had a word error rate of 7.2 percent in the Svarah column.
Checked against its source on October 17, 2026
In the source's own words
“Whisper large 1550M 7.2 1.6 2.7”
Svarah: Evaluating English ASR Systems on Indian Accents, Tahir Javed et al., arXiv, 2023, Table 1.
Open the original- What it measures
- Table 1 lists the Svarah column first, followed by the Speech Accent Archive native-speaker column (SAA_L1) and LibriSpeech clean (Libri). The 7.2 value is in the Svarah column.
- What it does not show
- One 2023 test with 117 speakers. Newer Whisper or commercial versions may score differently, and the models were run the way the authors set them up.