Fact · Languages and accents
32.7percent character error rate
In a 2024 study, the untuned Whisper-Tiny model had a character error rate of 32.7 percent on Japanese, and fine-tuning on Japanese data lowered it to 14.7 percent.
Checked against its source on October 17, 2026
In the source's own words
“Our results show that fine-tuning reduced Whisper-Tiny’s Character Error Rate (CER) from 32.7 to 20.8 with LoRA and to 14.7 with end-to-end fine-tuning”
Efficient Adaptation of Multilingual Models for Japanese ASR, Mark Bajo et al., arXiv, 2024, Abstract.
Open the original- What it measures
- Whisper-Tiny is the smallest Whisper model. The authors used Japanese-specific training data and two fine-tuning methods, LoRA and full end-to-end training.
- What it does not show
- Tiny models are far less accurate than the larger Whisper models. The figures show the gain from Japanese training, not the accuracy a user will get from the largest model.