KVoiceBench: Korean Spoken Question Answering Benchmark for SpeechLMs

KVoiceBench is a Korean spoken question answering benchmark constructed to evaluate speech language models. It was built by KRAFTON upon the English benchmark VoiceBench to address the lack of existing Korean evaluation resources. The dataset page was last updated on June 1, 2026.

Use Cases

• Benchmarking Korean speech model performance on spoken question answering tasks.
• Training or fine-tuning speech models for Korean language understanding.
• Comparing model capabilities across different languages using a VoiceBench-derived framework.
• Researching the transferability of speech model benchmarks from English to Korean.

Strengths

• Specifically addresses a gap in Korean speech model evaluation.
• Built upon the established English benchmark VoiceBench, suggesting a structured methodology.

Limitations

• Column-level documentation and sample data are unavailable, making data structure and content difficult to assess prior to download.
• The dataset size, row count, and file formats are unknown, limiting suitability assessment for specific tasks.
• Last updated 2026-06-01 05:22:12; freshness should be verified.

Provenance

Source

KRAFTON.

Collection Method

Constructed by building upon the English benchmark VoiceBench.

Freshness

Last updated 2026-06-01 05:22:12.

Geography

Korean language focus.