issai/WavCapsQA_Kazakh_Russian
收藏Hugging Face2026-04-29 更新2026-05-10 收录
下载链接:
https://hf-mirror.com/datasets/issai/WavCapsQA_Kazakh_Russian
下载链接
链接失效反馈官方服务:
资源简介:
该数据集是WavCaps-QA测试集的哈萨克语和俄语机器翻译版本,属于AudioBench基准测试套件的一部分,旨在评估音频语言模型在环境声音和音频场景中的问答能力。原始WavCaps-QA基准测试用于测试音频语言模型是否能回答关于非语音音频内容(包括环境声音、音乐、动物叫声、机械噪音和背景场景)的自由形式问题。此版本仅对文本字段(instruction, answer)进行了机器翻译,音频内容保持不变,因为音频内容是非语音的环境声音,无需重新合成。
This dataset is a machine-translated Kazakh and Russian adaptation of the WavCaps-QA test set from the AudioBench benchmark suite, designed for evaluating audio question answering over general environmental sounds and audio scenes in Audio-Language Models. The original WavCaps-QA benchmark probes whether audio-language models can answer free-form questions about non-speech audio content — including environmental sounds, music, animal vocalizations, mechanical noises, and ambient scenes. This release localizes the textual instructions and reference answers into Kazakh and Russian, enabling evaluation of multilingual audio-language models on the same audio stimuli. Audio is unchanged. Only the textual fields (`instruction`, `answer`) were machine-translated. The audio recordings are bit-identical copies of the originals — since the audio consists of non-speech environmental sounds, no re-synthesis was needed.
提供机构:
issai



