monster-monash/AudioMNIST

AudioMNIST数据集包含了60位不同年龄和性别说话者说出的数字0到9的音频录音，每位说话者每个数字有50个录音，总共30,000个单通道时间序列示例，每个示例大约1秒，采样频率为48 kHz。这些示例被分为10个类别，对应于数字0到9。数据集已根据说话者进行交叉验证分折处理，以确保训练和验证集的独立性。此外，还有一个降采样版本AudioMNIST-DS，时间序列长度为4,000。

The AudioMNIST dataset consists of audio recordings of 60 different speakers saying the digits 0 to 9, with 50 recordings per digit per speaker, totaling 30,000 univariate time series examples, each of length 47,998 (approximately 1 second of data sampled at 48 kHz), categorized into 10 classes representing the digits 0 to 9. The dataset has been split into cross-validation folds based on speaker to ensure that recordings from a given speaker do not appear in both the training and validation sets. There is also a downsampled version of the dataset called AudioMNIST-DS, where the time series are reduced to a length of 4,000.