EzAudioCaps

1. Prepare following dataset:

AudioSet: https://research.google.com/audioset/

AudioSet Strongly Labeled Subset: https://research.google.com/audioset/download_strong.html

VggSound: https://www.robots.ox.ac.uk/~vgg/data/vggsound/

AudioCaps: https://audiocaps.github.io/

2. The audio file name format:

AudioSet data: {ytb_id}\_{start_second\*1000}\_{end_second\*1000}

AudioSet Strongly Labeled Subset: {ytb_id}_{start_second\*1000} (10s clip from the the start second)

VggSound: {ytb_id}_{start_second} (10s clip from the the start second)

AudioCaps: {audiocap_id}

1. 准备如下数据集：
AudioSet：https://research.google.com/audioset/
AudioSet强标注子集（AudioSet Strongly Labeled Subset）：https://research.google.com/audioset/download_strong.html
VggSound：https://www.robots.ox.ac.uk/~vgg/data/vggsound/
AudioCaps：https://audiocaps.github.io/

2. 音频文件命名格式：
AudioSet数据集：{ytb_id}_{start_second*1000}_{end_second*1000}
AudioSet强标注子集：{ytb_id}_{start_second*1000}（取自起始秒的10秒音频片段）
VggSound：{ytb_id}_{start_second}（取自起始秒的10秒音频片段）
AudioCaps：{audiocap_id}

搜集汇总

背景与挑战

背景概述

EzAudioCaps是由OpenSound整合的音频数据集，基于AudioSet、VggSound和AudioCaps等多个来源构建。其音频文件命名遵循各源数据集的特定格式，例如使用YouTube ID和时间戳进行标识。