Fact · Languages and accents
4.0percent word error rate
On the FLEURS read-speech benchmark, Whisper large-v2 had a word error rate of 4.0 percent in Italian.
Checked against its source on October 17, 2026
In the source's own words
“Whisper large-v2 13.4 17.0 44.6 7.1 38.2 4.0 5.3”
Robust Speech Recognition via Large-Scale Weak Supervision (full text), Alec Radford et al., OpenAI (arXiv), 2022, Appendix D.2.4, FLEURS transcription table, Italian column.
Open the original (DOI 10.48550/arXiv.2212.04356)- What it measures
- Table D.2.4 of the appendix lists word error rates (%) on FLEURS. The header for this block reads Croatian, Hungarian, Armenian, Indonesian, Icelandic, Italian, Japanese, so the sixth figure is Italian.
- What it does not show
- FLEURS is read Wikipedia sentences, recorded in quiet rooms. It is a benchmark, not a measure of dictation at your desk, and it uses the paper's own text normalization.