Fact · Languages and accents
14.2percent mixed error rate
A 2023 adaptation of Whisper reached a 14.2 percent mixed error rate on Mandarin-English code-switched speech, training 5.6 percent additional parameters.
Checked against its source on October 17, 2026
In the source's own words
“achieves a 14.2% mixed error rate, surpassing state-of-the-art method, where only 5.6% additional parameters over Whisper are trained”
Attention-Guided Adaptation for Code-Switching Speech Recognition, Bobbi Aditya et al., arXiv, 2023, Abstract.
Open the original- What it measures
- The result uses Whisper-small with a corpus recorded in Singapore and Malaysia (SEAME). The authors trained the adapter on that corpus, so the number reflects in-domain training.
- What it does not show
- The result depends on training the adaptation on code-switched data. It is not what an unmodified Whisper model achieves.