Fact · Transcription
80.4percent word error rate (isiXhosa)
After fine-tuning Whisper medium on the same amount of transcribed child speech, the word error rate was 80.4 percent for isiXhosa, against 47.4 percent for Afrikaans.
Checked against its source on October 17, 2026
In the source's own words
“the WER for Afrikaans is significantly better than for isiXhosa: 47.4% compared to 80.4% for Whisper medium”
Speech Recognition for Automatically Assessing Afrikaans and isiXhosa Preschool Oral Narratives, Christiaan Jacobs et al., arXiv (Stellenbosch University and University of Edinburgh), 2025, Results section, first row of Table II.
Open the original- What it measures
- The models were fine-tuned on five minutes of transcribed speech from four- and five-year-old children telling stories. The authors attribute the gap partly to the mismatch with Whisper's pretraining languages, and they cite earlier work that finds agglutinative languages such as isiXhosa harder for ASR than Germanic languages such as Afrikaans.