Formula-SED
收藏arXiv2025-09-30 收录
下载链接：
https://yutoshibata07.github.io/Formula-SED/
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是通过数学公式生成的合成数据集，专门用于声音事件检测。它的设计旨在消除标签噪声和偏差，同时允许大规模的预训练，而无需承担数据收集成本，也无需担心隐私问题。该数据集包含一百万个样本，并已对50k、100k以及1M不同规模的样本进行了测试。其任务是声音事件检测（Sound Event Detection，简称Sed）。
This is a synthetic dataset generated via mathematical formulas, specifically designed for Sound Event Detection (SED). It is engineered to eliminate label noise and bias, while enabling large-scale pre-training without incurring data collection costs or privacy concerns. This dataset contains one million samples, and has been tested on subsets of 50k, 100k, and 1M samples respectively. The target task of this dataset is Sound Event Detection (SED).
提供机构：
Yuto Shibata et al.
