Fact · Languages and accents
31.11percent word error rate
On the CS-Dialogue Mandarin-English test set, Whisper Large-V2 had a 31.11 percent word error rate and a 15.29 percent mixed error rate, the highest error rates among the pre-trained models the authors tested.
Checked against its source on October 17, 2026
In the source's own words
“exhibits the highest error rates among the pre-trained models, particularly in WER (31.11%) and MER (15.29%)”
CS-Dialogue: A 104-Hour Dataset of Spontaneous Mandarin-English Code-Switching Dialogues for Speech Recognition, Jiaming Zhou et al., arXiv, 2025, Results, Table 8 and surrounding text.
Open the original- What it measures
- The other pre-trained models in the comparison were Qwen2-Audio, Paraformer and SenseVoice-Small.
- What it does not show
- Spontaneous dialogue in one dataset, compared with the authors' own choice of models. The numbers are a difficulty signal, not a leaderboard.