ASD

该数据集包含了来自88位阿拉伯语发言者的8800个发音，每位发言者对10个数字各发音十次。每个发音序列由13维的梅尔频率倒谱系数（MFCCs）组成，采样率为11,025Hz。数据集分为两个版本：数字版本（10个类别）和语音版本（88个类别）。训练集包含6600个样本，测试集包含2200个样本，且训练集与测试集中的发言者没有重叠。规模上，该数据集共有8800个发音，任务是对语音和数字进行分类。

This dataset contains 8800 utterances from 88 Arabic speakers, where each speaker pronounces each of the 10 target digits ten times. Each utterance is represented by 13-dimensional Mel-Frequency Cepstral Coefficients (MFCCs), with a sampling rate of 11,025 Hz. The dataset offers two variants: the digit classification variant (10 classes) and the speaker classification variant (88 classes). The training set consists of 6600 samples, and the test set contains 2200 samples, with no overlapping speakers between the training and test partitions. In terms of scale, the dataset totals 8800 utterances, and the core tasks are digit classification and speaker classification.