lighthouse-emnlp2024/AudioMomentRetrievalFromLongAudio_DCASE2026EvaluationData
收藏资源简介:
该数据集为DCASE Challenge 2026 Task 6提供支持,包含使用CLAP模型和滑动窗口提取的音频与文本特征,特征提取协议与CASTELLA数据集保持一致。数据集包含clap和clap_text两个目录,分别存储音频特征文件(如dcase2026_evaluation_audio_{vid}.npz)和文本特征文件(如qiddcase2026_evaluation_q{qid}.npz),用于模型训练和评估。数据对应提交模板中的qid和vid,旨在支持相关时刻预测任务。
This dataset supports DCASE Challenge 2026 Task 6. It includes audio and text features extracted using the CLAP model with a sliding window, and the feature extraction protocol is consistent with that of the CASTELLA dataset. The dataset contains two directories: clap and clap_text, which store audio feature files (e.g., dcase2026_evaluation_audio_{vid}.npz) and text feature files (e.g., qiddcase2026_evaluation_q{qid}.npz) respectively for model training and evaluation. The data corresponds to the qid and vid specified in the submission template, and is intended to support the relevant moment prediction task.




