Preprint
Efficient Adaptation of Multilingual Models for Japanese ASR
Mark Bajo et al., arXiv, 2024
Read the originalRead on October 17, 2026
- What it is
- A study of adapting small multilingual Whisper models to Japanese speech. It reports character error rate for Whisper-Tiny and Whisper-Base before and after fine-tuning on Japanese data, and it explains why CER is used for Japanese.
- How much weight it carries
- Preprint with a clear method and public results. The test sets are specific to the study and small models are tested, so the numbers show relative gains, not how Whisper large models perform in ordinary dictation.
Facts taken from this source
- 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.
“surpassing Whisper-Base’s CER of 20.2.” (Abstract and results)
- 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.
“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” (Abstract)