Preprint

First Steps Towards Voice Anonymization for Code-Switching Speech

Sarina Meyer et al., arXiv, 2025

Read on October 17, 2026

What it is: A 2025 study that tested Whisper on two conversational code-switching datasets, Mandarin-English and Spanish-English. It reports the error rate on code-switched utterances and describes how Whisper's transcripts change the speakers' code-switching.
How much weight it carries: A preprint that has not been checked by us against a journal version. The data are spontaneous conversational speech with modest audio quality, so the figures describe these two corpora and one Whisper setup, not code-switching in general. The paper's main topic is voice anonymization, so its speech recognition numbers are secondary.

Facts taken from this source

The authors found that speech recognition with Whisper significantly altered and reduced the speakers' original code-switching.
“the original code-switching behavior is significantly altered and reduced during speech recognition” (Results discussion)
On two conversational code-switching corpora, Whisper's mixed error rate on code-switched utterances was about 39 percent.
“39% MER for code-switched utterances and find that the original code-switching behavior is significantly altered and reduced during speech recognition” (Results on the two code-switching corpora)