Fact · Languages and accents
39percent mixed error rate
On two conversational code-switching corpora, Whisper's mixed error rate on code-switched utterances was about 39 percent.
Checked against its source on October 17, 2026
In the source's own words
“39% MER for code-switched utterances and find that the original code-switching behavior is significantly altered and reduced during speech recognition”
First Steps Towards Voice Anonymization for Code-Switching Speech, Sarina Meyer et al., arXiv, 2025, Results on the two code-switching corpora.
Open the original- What it measures
- Mixed error rate counts errors in both languages. The same paper reports a higher 20 to 30 percent mixed error rate on monolingual utterances of these corpora.
- What it does not show
- The figure is for spontaneous conversational speech with modest audio quality, one Whisper setup and two language pairs. It is not a general error rate for code-switching.
Source
Cited on
- Multilingual Dictation: Two Languages, Auto-Detect, MixingGuide
- Whisper Supported Languages and Accuracy by LanguageGuide
- Can speech to text handle Hinglish, Spanglish and other mixed languages?Question
- Can you dictate in two languages at the same time?Question
- Why does Whisper transcribe in the wrong language?Question
- Code-switchingGlossary