mteb/urbansound8K
收藏资源简介:
UrbanSound8K数据集是一个用于音频分类的数据集,包含8732个标记的城市声音片段(<=4秒),分为10个类别:空调声、汽车喇叭声、儿童玩耍声、狗叫声、钻孔声、发动机怠速声、枪声、破碎锤声、警笛声和街头音乐声。这些类别来自城市声音分类法。所有片段均来自上传至Freesound.org的现场录音。文件已预先分为十个折叠(fold1-fold10),以便于复现和与文献中报告的自动分类结果进行比较。除了声音片段外,还提供了一个包含每个片段元数据的CSV文件。数据集中的音频文件为WAV格式,采样率、位深度和通道数与原始文件相同(可能因文件而异)。元数据文件包括音频文件名、Freesound ID、开始和结束时间、显著性评级、折叠编号、类别ID和类别名称等信息。为避免常见错误,README强调不要重新洗牌数据,应使用预定义的10折交叉验证,并报告平均分数以确保结果的可比性。数据集来源于Freesound.org的现场录音,用于城市声音研究。
The UrbanSound8K dataset is an audio classification dataset containing 8732 labeled sound excerpts (<=4s) of urban sounds from 10 classes: air_conditioner, car_horn, children_playing, dog_bark, drilling, engine_idling, gun_shot, jackhammer, siren, and street_music. The classes are drawn from the urban sound taxonomy. All excerpts are taken from field recordings uploaded to Freesound.org. The files are pre-sorted into ten folds (fold1-fold10) to aid in the reproduction of and comparison with prior automatic classification results. In addition to the sound excerpts, a CSV file containing metadata for each excerpt is provided. The audio files are in WAV format, with sampling rate, bit depth, and number of channels matching the original files (which may vary). The metadata includes the audio file name, Freesound ID, start and end times, salience rating, fold number, class ID, and class name. The README cautions against common pitfalls such as reshuffling the data or evaluating on a single split, recommending the use of predefined 10-fold cross-validation and reporting average scores for comparable results. The dataset is sourced from Freesound.org field recordings and is intended for urban sound research.




