DCASE2025 Task 4 Dataset
收藏资源简介:
DCASE2025 Task 4 数据集是为DCASE 2025挑战赛中的空间语义分割声音场景(S5)任务而创建的,旨在从多通道空间输入信号中检测和分离声音事件。该数据集包括孤立的声音事件、房间脉冲响应、环境噪声和干扰声音,所有这些数据都是为新任务而重新录制的。它用于训练和评估沉浸式通信技术系统,包括扩展现实(XR)。数据集共包含18个类别的声音事件,每个音频片段长度固定为10秒,包含1到3个同时发生的声音事件。数据集的开发集包含训练、验证和测试三个子集,而评估集则是全新录制的,不包含任何公开可用的数据。
The DCASE2025 Task 4 Dataset was created for the Spatial Semantic Segmentation of Sound Scenes (S5) task in the DCASE 2025 Challenge, aiming to detect and separate sound events from multi-channel spatial input signals. This dataset encompasses isolated sound events, room impulse responses, ambient noise and interfering sounds, all of which were re-recorded specifically for this new task. It is used for training and evaluating immersive communication technology systems, including extended reality (XR). The dataset consists of 18 categories of sound events, with each audio clip having a fixed duration of 10 seconds and containing 1 to 3 simultaneously occurring sound events. The development set includes three subsets: training, validation and test, while the evaluation set is newly recorded and does not contain any publicly available data.

- 1Description and Discussion on DCASE 2025 Challenge Task 4: Spatial Semantic Segmentation of Sound ScenesNTT Corporation, Japan · 2025年



