LASS-Sub-Test
收藏arXiv2025-09-30 收录
下载链接：
https://liuxubo717.github.io/LASS-demopage
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是从LASS-Test中选取的子集，包含了250个测试音频混合样本，每个样本都附有六个标注的标题。此外，这个子集中的每个音频都有多个由人类标注的标题，这为使用多种描述进行评估提供了可能。该数据集的规模为250个测试混合样本，其任务是语言查询的音频源分离。
This dataset is a subset curated from LASS-Test, comprising 250 test audio mixture samples. Each sample is paired with six annotated titles. Furthermore, each audio within this subset features multiple human-annotated titles, which facilitates evaluation using varied descriptive inputs. The core task of this dataset is audio source separation conditioned on natural language queries.
提供机构：
Based on AudioCaps dataset
