URBAN-SED

URBAN-SED 是一个包含 10,000 个音景的数据集，其中包含使用刮板库生成的声音事件注释。该数据集包括 10,000 个音景，总计近 30 小时，包括近 50,000 个带注释的声音事件。每个音景的长度为 10 秒，并且具有类似于在城市环境中经常听到的典型“嗡嗡声”的布朗噪声背景。每个音景包含来自以下类别的 1-9 个声音事件：air_conditioner、car_horn、children_playing、dog_bark、drilling、engine_idling、gun_shot、jackhammer、siren 和 street_music。声音事件的源材料是来自 UrbanSound8K 数据集的剪辑。 URBAN-SED 预先分为三组：训练、验证和测试。训练集中有 6000 个音景，使用 UrbanSound8K 中第 1-6 折的剪辑生成，验证集中有 2000 个音景，使用 UrbanSound8K 中第 7-8 折的剪辑生成，测试集中有 2000 个音景，使用来自在 UrbanSound8K 中折叠 9-10。URBAN-SED is a dataset containing 10,000 soundscapes with sound event annotations generated using the Scaper library. This dataset comprises 10,000 soundscapes totaling nearly 30 hours of audio, with approximately 50,000 annotated sound events in total. Each soundscape has a duration of 10 seconds and features a Brown noise background similar to the typical "hum" often heard in urban environments. Each soundscape contains 1 to 9 sound events belonging to the following categories: air_conditioner, car_horn, children_playing, dog_bark, drilling, engine_idling, gun_shot, jackhammer, siren, and street_music. The source material for these sound events is audio clips sourced from the UrbanSound8K dataset. URBAN-SED is pre-partitioned into three subsets: training, validation, and test. The training subset contains 6,000 soundscapes generated using clips from folds 1–6 of UrbanSound8K; the validation subset includes 2,000 soundscapes generated using clips from folds 7–8 of UrbanSound8K; and the test subset has 2,000 soundscapes generated using clips from folds 9–10 of UrbanSound8K.

搜集汇总

背景与挑战

背景概述

URBAN-SED是一个包含10,000个城市环境音景的数据集，总计近30小时音频，标注了近50,000个声音事件，覆盖10种常见城市声音类别。数据集已预先分为训练、验证和测试集，适用于声音事件检测任务。