ASVspoof5 Development Set

该数据集包含了八种欺骗方法（A09-A16），其中真实样本有23,645个，伪造样本有82,588个，总计音频样本数为106,233个。此外，该开发集用于验证在训练集上训练的模型的性能。规模上，数据集包含了106,233个音频样本，任务是对音频深度伪造进行检测。

This dataset includes eight spoofing methods (A09-A16). It contains 23,645 bona fide audio samples and 82,588 spoofed audio samples, with a total of 106,233 audio samples. Additionally, this dataset serves as the development set for validating the performance of models trained on the training set, with its core task being audio deepfake detection.

搜集汇总

背景与挑战

背景概述

ASVspoof5 Development Set是用于语音深度伪造检测的开发数据集，允许使用外部数据和预训练模型，评估指标包括minDCF、actDCF、Cllr和EER。竞赛分为多个阶段，每个阶段有不同的提交限制。