issai/SAKURA_Kazakh_Russian

该数据集名为SAKURA — Kazakh & Russian，是SAKURA基准的机器翻译哈萨克语和俄语版本，用于评估音频语言模型中的单跳和多跳推理能力。数据集包含四个子集：AnimalQA、EmotionQA、GenderQA和LanguageQA，每个子集探究不同的感知属性。数据集结构包括音频录音和翻译成哈萨克语和俄语的文本字段，音频部分与原始版本保持不变。该数据集旨在使用相同的音频刺激但本地化的文本指令和答案来评估多语言音频语言模型。

This dataset, named SAKURA — Kazakh & Russian, is a machine-translated Kazakh and Russian adaptation of the SAKURA benchmark for evaluating single-hop and multi-hop reasoning in Audio-Language Models. The dataset includes four subsets: AnimalQA, EmotionQA, GenderQA, and LanguageQA, each probing different perceptual attributes. The dataset structure consists of audio recordings and textual fields translated into Kazakh and Russian, with the audio remaining unchanged from the original. The dataset is designed to evaluate multilingual audio-language models using the same audio stimuli but with localized textual instructions and answers.