Fact · Transcription
38.3percent word error rate
On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 38.3 percent for Marathi.
Checked against its source on October 17, 2026
In the source's own words
“Model Lingala Lao Lithuanian Latvian Maori Macedonian Malayalam Mongolian Marathi Malay ... Whisper large-v2 75.6 101.5 28.1 23.1 38.5 16.5 100.7 110.5 38.3 8.7”
Robust Speech Recognition via Large-Scale Weak Supervision (full text), Alec Radford et al., OpenAI (arXiv), 2022, Table 13 (Fleurs), column Marathi.
Open the original (DOI 10.48550/arXiv.2212.04356)- What it measures
- Fleurs is a read-speech benchmark in about 100 languages. The figure is for the largest Whisper model in the table, large-v2, from 2022.
- What it does not show
- Read sentences, not conversation. The 2022 table is older than the newer comparisons in our guide, so newer systems may differ.