DFADD: The Diffusion and Flow-Matching Based Audio Deepfake Dataset

TL;DR

This paper introduces the DFADD dataset, a collection of deepfake audio generated by diffusion and flow-matching TTS models, highlighting the need for more robust anti-spoofing methods against advanced synthetic speech.

Contribution

The paper presents the DFADD dataset, specifically designed to evaluate and improve anti-spoofing models against diffusion and flow-matching based audio deepfakes.

Findings

Current anti-spoofing models lack robustness against diffusion and flow-matching generated audio.

DFADD dataset provides a new benchmark for evaluating anti-spoofing techniques.

Advanced TTS models produce highly human-like audio that challenges existing detection methods.

Abstract

Mainstream zero-shot TTS production systems like Voicebox and Seed-TTS achieve human parity speech by leveraging Flow-matching and Diffusion models, respectively. Unfortunately, human-level audio synthesis leads to identity misuse and information security issues. Currently, many antispoofing models have been developed against deepfake audio. However, the efficacy of current state-of-the-art anti-spoofing models in countering audio synthesized by diffusion and flowmatching based TTS systems remains unknown. In this paper, we proposed the Diffusion and Flow-matching based Audio Deepfake (DFADD) dataset. The DFADD dataset collected the deepfake audio based on advanced diffusion and flowmatching TTS models. Additionally, we reveal that current anti-spoofing models lack sufficient robustness against highly human-like audio generated by diffusion and flow-matching TTS systems. The proposed DFADD dataset addresses this gap and provides a valuable resource for developing more resilient anti-spoofing models.