SpatialTAS

该数据集名为SpatialTAS，包含了大规模的模拟双耳音频，这些音频具有对声源位置详尽且灵活的文字描述，旨在提升增强现实（AR）、虚拟现实（VR）以及具身人工智能应用中的沉浸式体验。数据集分为376,104个训练样本、732个验证样本以及4,000个测试样本，其中训练样本涵盖了多个类别。规模上，共有376,104个样本（其中376,000个为模拟双耳音频样本）。该数据集的任务包括音频空间化以及对空间语义一致性的评估。

This dataset, named SpatialTAS, contains large-scale simulated binaural audio with detailed and flexible textual descriptions of sound source positions, aiming to enhance the immersive experience in augmented reality (AR), virtual reality (VR), and embodied artificial intelligence applications. The dataset is split into 376,104 training samples, 732 validation samples, and 4,000 test samples, with the training samples covering multiple categories. In terms of scale, there are a total of 376,104 samples, among which 376,000 are simulated binaural audio samples. The tasks of this dataset include audio spatialization and the evaluation of spatial semantic consistency.