DCASE 2018 Task 4
收藏资源简介:
DCASE2018 Task 4 是用于家庭环境中大规模弱标记半监督声音事件检测的数据集。这些数据是 YouTube 视频摘录,侧重于家庭环境,可用于环境辅助生活应用。选择该领域是因为科学挑战(各种各样的声音、时间本地化的事件......)和潜在的工业应用。具体来说,该任务使用了 Google 的“音频集:音频事件的本体和人工标记数据集”的一个子集。 Audioset 由 632 个声音事件类的扩展本体和从 200 万个 Youtube 视频中提取的 200 万个人工标记的 10 秒声音片段(不到 21% 短于 10 秒)的集合组成。本体被指定为事件类别的层次图,涵盖广泛的人类和动物声音、乐器和流派以及常见的日常环境声音。任务 4 侧重于 Audioset 的一个子集,该子集由 10 类声音事件组成:语音、狗、猫、警铃响起、盘子、油炸、搅拌机、自来水、吸尘器、电动剃须刀牙刷。
DCASE2018 Task 4 is a dataset for large-scale weakly-labeled semi-supervised sound event detection in home environments. The data consists of excerpts from YouTube videos, focusing on home environments, and can be applied to ambient assisted living applications. This domain was selected due to both its scientific challenges (e.g., diverse sound types, temporally localized events, etc.) and potential industrial applications. Specifically, this task uses a subset of Google's "AudioSet: An Ontology and Human-labeled Dataset of Audio Events". AudioSet comprises an extended ontology of 632 sound event classes and a collection of 2 million manually labeled 10-second audio clips extracted from 2 million YouTube videos, with less than 21% of the clips being shorter than 10 seconds. The ontology is defined as a hierarchical graph of event categories, covering a wide range of human and animal sounds, musical instruments and genres, as well as common daily environmental sounds. Task 4 focuses on a subset of AudioSet that includes 10 sound event classes: speech, dog, cat, alarm bell ringing, dishes, frying, blender, running water, vacuum cleaner, and electric shaver/toothbrush.

- DCASE 2018 Task 4首次发表,专注于大规模机器聆听挑战,旨在评估和提升音频事件检测与定位的算法性能。
- DCASE 2018 Task 4首次应用于实际场景,通过提供多源音频数据集,推动了音频事件检测技术的发展。



