Korean-Japanese-Code-Switching-Speech

This dataset contains Korean-Japanese code-switching speech recordings with sentence-level transcriptions. It was introduced in the paper Towards Truly Multilingual ASR: Generalizing Code-Switching ASR to Unseen Language Pairs.

Since there is an extremely small amount of Korean-Japanese code-switching data available, this dataset was designed to be used as a small-scale evaluation dataset.

The dataset consists of code-switching recordings produced by two Korean participants who are fluent in both Korean and Japanese. It can be used to evaluate the basic Korean-Japanese code-switching capability of ASR models, but it may not be suitable for evaluations that require diverse speaker characteristics such as nationality, gender, and age group.

Dataset Summary

• Split: test
• Rows: 450
• Audio format: 16 kHz WAV stored in the Hugging Face Audio feature
• Total duration: 2902.89 seconds
• Main-language labels: Korean, Japanese
• Code-switching level labels: word, phrase, sentence

Columns

• audio: speech audio
• text: transcription
• script_id: source script identifier
• main_language: primary language of the utterance
• cs_level: code-switching granularity
• category: topic/category label
• duration_sec: audio duration in seconds

How to Cite

@misc{paik2026towards,
      title={Towards Truly Multilingual ASR: Generalizing Code-Switching ASR to Unseen Language Pairs}, 
      author={Gio Paik and Hyunseo Shin and Soungmin Lee},
      year={2026},
      eprint={2606.05846},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2606.05846}, 
}