Contributors：Wenchang Cao, Wei Xie, Yanxiong Li  

    Three audio datasets, including FSC-89, NSynth-100 and LS-100, are constructed as experimental data to evaluate the performance of audio classification methods. The details of these three audio datasets are described at three websites [1]-[3]. What’s more, they can be downloaded from the above three websites and be freely used for research purpose.

    The dataset of FSC-89 has 89 sound events (audio classes) which cover a diverse range of real-world sounds, from human andanimal sounds to natural, musical or miscellaneous sounds. The dataset of NSynth-100has 100 sounds (musical notes) of different musical instruments. The dataset of LS-100 consists of speech samples uttered by 100 different speakers (classes).

[1] https://www.modelscope.cn/datasets/pp199124903/LS-100/summary

[2] https://www.modelscope.cn/datasets/pp199124903/FSC-89/summary

[3] https://www.modelscope.cn/datasets/pp199124903/NSynth-100/summary