--- dataset_info: - config_name: advbench-test features: - name: id dtype: string - name: audio dtype: audio - name: transcription dtype: string - name: answer dtype: string - name: capability dtype: string splits: - name: test num_examples: 513 - config_name: alpacaeval_full-test features: - name: id dtype: string - name: audio dtype: audio - name: transcription dtype: string - name: answer dtype: string - name: capability dtype: string splits: - name: test num_examples: 614 - config_name: bbh-test features: - name: id dtype: string - name: audio dtype: audio - name: transcription dtype: string - name: answer dtype: string - name: capability dtype: string splits: - name: test num_examples: 750 - config_name: commoneval-test features: - name: id dtype: string - name: audio dtype: audio - name: transcription dtype: string - name: answer dtype: string - name: capability dtype: string splits: - name: test num_examples: 194 - config_name: ifeval-test features: - name: id dtype: string - name: audio dtype: audio - name: transcription dtype: string - name: answer dtype: string - name: capability dtype: string splits: - name: test num_examples: 279 - config_name: mmsu features: - name: id dtype: string - name: audio dtype: audio - name: transcription dtype: string - name: answer dtype: string - name: capability dtype: string splits: - name: test num_examples: 3006 - config_name: openbookqa-test features: - name: id dtype: string - name: audio dtype: audio - name: transcription dtype: string - name: answer dtype: string - name: capability dtype: string splits: - name: test num_examples: 445 - config_name: sd-qa features: - name: id dtype: string - name: audio dtype: audio - name: transcription dtype: string - name: answer dtype: string - name: capability dtype: string splits: - name: test num_examples: 533 - config_name: wildvoice-test features: - name: id dtype: string - name: audio dtype: audio - name: transcription dtype: string - name: answer dtype: string - name: capability dtype: string splits: - name: test num_examples: 972 configs: - config_name: advbench-test data_files: - split: test path: advbench-test/test-* - config_name: alpacaeval_full-test data_files: - split: test path: alpacaeval_full-test/test-* - config_name: bbh-test data_files: - split: test path: bbh-test/test-* - config_name: commoneval-test data_files: - split: test path: commoneval-test/test-* - config_name: ifeval-test data_files: - split: test path: ifeval-test/test-* - config_name: mmsu data_files: - split: test path: mmsu/test-* - config_name: openbookqa-test data_files: - split: test path: openbookqa-test/test-* - config_name: sd-qa data_files: - split: test path: sd-qa/test-* - config_name: wildvoice-test data_files: - split: test path: wildvoice-test/test-* license: apache-2.0 task_categories: - automatic-speech-recognition - question-answering language: - ko pretty_name: KVoiceBench size_categories: - 1K Homepage Hugging Face License

## Evaluation Evaluation code for KVoiceBench, KOpenAudioBench, and KMMAU is available in [Raon-Eval](https://github.com/krafton-ai/Raon-Eval). **KVoiceBench** is a Korean spoken question answering benchmark for evaluating speech models, built upon [VoiceBench](https://arxiv.org/abs/2410.17196). ## Dataset Description As there is no existing benchmark for evaluating Korean spoken question answering capabilities, we construct KVoiceBench by building upon the English benchmark VoiceBench. ## Construction Process Following the paper's SpokenQA transfer framework, KVoiceBench was built by transferring VoiceBench into Korean. Source samples with deterministic answers were first audited through a reviewer/meta-reviewer LLM process; confirmed label errors were corrected and invalid cases were excluded. The retained items were then hypertranslated with a human-agent rulebook. A separate human-agent normalization loop audited the hypertranslated files, organized likely TTS failure cases into a normalization rulebook, and produced speech-friendly Korean before Qwen3-TTS synthesis with Korean reference voices and WER-based resynthesis checks. ## Subsets The dataset consists of **7,306** samples across 9 subsets: | Subset | Samples | Task Type | |---|---|---| | KOpenBookQA | 445 | Multiple-choice QA | | KMMSU | 3,006 | Multiple-choice QA | | KBBH | 750 | Binary reasoning | | KSD-QA | 533 | Short-answer QA | | KAlpacaEval | 614 | Open-ended | | KCommonEval | 194 | Open-ended | | KWildVoice | 972 | Open-ended | | KIFEval | 279 | Instruction following | | KAdvBench | 513 | Safety | ## Features - **id** (string): Unique sample identifier - **audio** (Audio): Synthesized Korean speech audio - **transcription** (string): Korean text transcription of the audio - **answer** (string): Reference answer (empty string for open-ended/safety subsets) - **capability** (string): Evaluated capability category ## Usage ```python from datasets import load_dataset # Load a specific subset ds = load_dataset("KRAFTON/KVoiceBench", "bbh-test", split="test") print(ds[0]) # {'id': 'bbh_navigate_147', 'audio': {'path': ..., 'array': array([...]), 'sampling_rate': 24000}, # 'transcription': '...', 'answer': '...', 'capability': 'context_related/multi_hop_reasoning'} # Load all subsets subsets = [ "advbench-test", "alpacaeval_full-test", "bbh-test", "commoneval-test", "ifeval-test", "mmsu", "openbookqa-test", "sd-qa", "wildvoice-test" ] for subset in subsets: ds = load_dataset("KRAFTON/KVoiceBench", subset, split="test") print(f"{subset}: {len(ds)} samples") ``` ## Citation ```bibtex @misc{kim2026kvoicebenchkopenaudiobenchkmmau, title={KVoiceBench, KOpenAudioBench, and KMMAU: Agent-Driven Korean Speech Benchmarks for Evaluating SpeechLMs}, author={Kim, Haechan and Chung, Seungjun and Park, Inkyu and Lee, Jihoo and Lee, Jonghyun}, year={2026}, eprint={2605.27984}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/pdf/2605.27984v1} } ``` ## License This dataset is released under the [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0). © 2026 KRAFTON