This dataset contains the audio and the single/multi-hop questions/answers of the language track of the SAKURA benchmark from Interspeech 2025 paper, "SAKURA: On the Multi-hop Reasoning of Large Audio-Language Models Based on Speech and Audio Information".
The fields of the dataset are:
file: The filename of the audio files.
audio: The audio recordings.
attribute_label: The attribute labels (i.e., the language spoken in the speech) of the audio… See the full description on the dataset page:
https://huggingface.co/datasets/SLLM-multi-hop/LanguageQA.
