DCASE2020 Acoustic Scene Classification (ASC) Task

该数据集专为声学场景分类任务而设计，包含了由多种设备录制的音频片段。其中，训练数据来源于一个源设备，同时还包括了来自八个目标设备的配对音频片段。此外，该数据集还包含了从每段音频中提取的对数梅尔滤波器组（LMFB）特征，这些特征在分类前被缩放到[0,1]的范围，以解决声学分类中的设备适应性问题。规模上，该数据集大约包含了来自源设备的10,000个片段，以及每个目标设备的750个片段。其任务是对声学场景进行分类。

This dataset is specifically designed for acoustic scene classification tasks, and comprises audio clips recorded using multiple devices. The training data includes audio clips from one source device, alongside paired audio clips from eight target devices. Additionally, the dataset contains log Mel filter bank (LMFB) features extracted from each audio clip, which are scaled to the range [0, 1] prior to classification to address the device adaptation issue in acoustic classification. In terms of scale, the dataset contains approximately 10,000 audio clips from the source device, and 750 clips for each target device. The task of this dataset is acoustic scene classification.