MBZUAI/NADI-2025-Sub-task-3-all

该数据集是一个包含多种阿拉伯语方言和录音条件的语音和转录数据集，适用于闭轨训练和模型开发。它包括方言、现代标准、古典和代码转换的阿拉伯语，共有超过85K的训练句子。数据集的所有子集（除了Mixat和ArzEn）都带有标点。经过处理，数据集包含57K的训练样本和1.5K的开发样本。

This dataset is a collection of Arabic speech and transcription with a variety of dialects and recording conditions, suitable for closed track training and model development. It includes dialectal, modern standard, classical, and code-switched Arabic, with a total of over 85K training sentences. All subsets of the dataset are diacritized except for Mixat and ArzEn. After processing, the dataset contains 57K training samples and 1.5K development samples.