ADD 2022: the First Audio Deep Synthesis Detection Challenge

TL;DR

ADD 2022 introduces the first Audio Deep Synthesis Detection Challenge to address realistic deepfake scenarios not fully covered by prior challenges. It defines three tracks—LF (low-quality fake), PF (partially fake), and FG (audio fake game with generation and detection tasks)—with shared training/dev data and track-specific adaptation/test sets based on the AISHELL-3 Mandarin corpus. Evaluation uses threshold-free equal error rate for detection tasks and deception metrics for adversarial generation, plus a weighted WEER ranking across two evaluation rounds in track 3.2, and reports baseline results from six detectors (GMM, LCNN, RawNet2) showing substantial remaining gaps and generalization challenges. The findings underscore the need for robust, generalizable defenses against diverse audio deepfakes and motivate future work on metric design and dataset construction for real-world deployment.

Abstract

Audio deepfake detection is an emerging topic, which was included in the ASVspoof 2021. However, the recent shared tasks have not covered many real-life and challenging scenarios. The first Audio Deep synthesis Detection challenge (ADD) was motivated to fill in the gap. The ADD 2022 includes three tracks: low-quality fake audio detection (LF), partially fake audio detection (PF) and audio fake game (FG). The LF track focuses on dealing with bona fide and fully fake utterances with various real-world noises etc. The PF track aims to distinguish the partially fake audio from the real. The FG track is a rivalry game, which includes two tasks: an audio generation task and an audio fake detection task. In this paper, we describe the datasets, evaluation metrics, and protocols. We also report major findings that reflect the recent advances in audio deepfake detection tasks.