DESED (Domestic environment sound event detection)

DESED 数据集是一个旨在识别家庭环境中的声音事件类别的数据集。

该数据集旨在用于声音事件检测（SED，识别具有时间边界的事件），但它也可用于音频标记（AT，指示音频文件中存在事件）。

目前，该数据集由 10 个事件类组成，可在 10 秒的音频文件中进行识别。

课程：闹钟/铃声/铃声，搅拌机，猫，狗，盘子，

电动剃须刀/牙刷、油炸、自来水、语音、吸尘器。

The DESED dataset is a collection designed to recognize sound event categories in household environments.

This dataset is primarily intended for sound event detection (SED, the task of identifying events with explicit temporal boundaries), but it can also be applied to audio tagging (AT, the task of indicating the presence of events within audio files).

Currently, this dataset comprises 10 event classes, which can be recognized in 10-second audio clips.

The event classes are as follows: alarm clock/bell/ringtone, blender, cat, dog, dishes, electric shaver/toothbrush, frying, running water, speech, vacuum cleaner.

搜集汇总

背景与挑战

背景概述

DESED数据集专注于家庭环境中的声音事件检测，包含10个常见事件类别，如闹钟、猫和狗等，适用于声音识别和时间边界分析。该数据集由多所研究机构于2019年联合发布，旨在支持弱标签数据和声音合成的研究。