ArlingtonCL2/Dog-Vocal-Separation

这是一个狗叫声分离的音频数据集，包含训练集、验证集和测试集。训练集提供了混合音频及其对应的10秒狗叫声真实音频，而测试集仅包含混合音频。数据集中的音频采样率为32,000 kHz，且狗叫声经过填充至10秒长度，并与AudioSet中的背景噪音混合。数据集的总时长约为348小时（训练集）、46小时（验证集）和8小时（测试集）。This is an audio dataset for dog vocal separation, containing training, validation, and test sets. The training set provides mixed audio and its corresponding 10-second ground truth dog vocal audio, while the test set only includes mixed audio. The audio in the dataset is sampled at 32,000 kHz, and the dog vocals are padded to 10 seconds in length and mixed with background noise from AudioSet. The total duration of the dataset is approximately 348 hours (training set), 46 hours (validation set), and 8 hours (test set).

搜集汇总

背景与挑战

背景概述

该数据集名为'Dog Vocal Separation'，是一个用于狗叫声分离的音频数据集，包含训练、验证和测试集，总大小180 GB，音频采样率为32,000 kHz。数据集提供狗叫声与背景噪音的混合音频及其对应的纯净狗叫声作为真值，旨在支持音频分离任务，并用于IJCAI-2025挑战赛，评估指标为SI-SDR。