Fact · Languages and accents
20.2percent character error rate
In a 2024 study of Japanese speech recognition, the original Whisper-Base model had a character error rate of 20.2 percent on the study's Japanese test data.
Checked against its source on October 17, 2026
In the source's own words
“surpassing Whisper-Base’s CER of 20.2.”
Efficient Adaptation of Multilingual Models for Japanese ASR, Mark Bajo et al., arXiv, 2024, Abstract and results.
Open the original- What it measures
- The study fine-tuned the small Whisper-Tiny model on Japanese data and compared it with the untuned Whisper-Base. It measured both word error rate and character error rate and used character error rate as the main Japanese result.
- What it does not show
- Whisper-Base is one of the smallest Whisper models. The score is for the study's own Japanese test sets, so it does not describe the larger Whisper models or everyday dictation.