Jinbo-HU/PSELDNets

DataSynthSELD数据集是一个用于声音事件定位和检测的大规模合成数据集。它包含67,000个1分钟的音频片段用于训练，总计约1,117小时，以及3,060个1分钟的音频片段用于测试，总计约51小时。数据集的特征是包含170种声音类别，并通过将FSD50K数据集中的声音事件片段与模拟的SRIRs（用于训练）或从TAU-SRIR DB收集的SRIRs（用于测试）进行卷积生成。数据集生成工具为SELD-Data-Generator。

The DataSynthSELD dataset is a large-scale synthetic dataset for sound event localization and detection. It contains 67,000 1-minute clips for training, amounting to approximately 1,117 hours, and 3,060 1-minute clips for testing, amounting to roughly 51 hours. The dataset features an ontology of 170 sound classes and is generated by convolving sound event clips from the FSD50K dataset with simulated SRIRs (for training) or collected SRIRs from the TAU-SRIR DB (for testing). The dataset is generated using the SELD-Data-Generator tool.