DCASE 2024 Sound Scene Synthesis Dataset

该数据集包含了310个音频标注样本，其中60个用于开发，250个用于评估，旨在针对高质量的声响场景匹配特定提示。此外，音频样本来源于Freesound.org以及私人图书馆，被分类为前景和背景声音，这有助于创建逼真的声响场景。总的来说，该数据集的规模为310个音频标注，任务聚焦于声响场景的合成。This dataset contains 310 annotated audio samples, with 60 reserved for development and 250 for evaluation, aimed at matching specific prompts to generate high-quality acoustic scenes. Additionally, the audio samples are sourced from Freesound.org and private audio libraries, and are categorized into foreground and background sounds, which facilitates the creation of realistic acoustic scenes. Overall, this dataset comprises 310 annotated audio samples, with its core task focusing on acoustic scene synthesis.

搜集汇总

背景与挑战

背景概述

该数据集是DCASE 2024挑战赛Task7（声音场景合成）的配套数据，主要用于评估文本到音频生成任务。它作为Frechet Audio Distance工具库的基准参考集，包含音频嵌入文件和描述文本，但详情页未提供数据规模、具体样本或采集方式等细节。