Diffuse or Confuse: A Diffusion Deepfake Speech Dataset

TL;DR

This paper introduces a new diffusion-based synthetic speech dataset, evaluates its quality and detectability, and compares it to traditional methods, highlighting its potential impact on deepfake detection systems.

Contribution

It presents a diffusion model-based speech dataset and assesses its realism and detection challenges compared to non-diffusion deepfakes.

Findings

Detection accuracy for diffusion deepfakes is similar to non-diffusion deepfakes.

Diffusion vocoders cause minimal re-vocoding impact on speech quality.

Diffusion deepfakes pose comparable threats to detection systems.

Abstract

Advancements in artificial intelligence and machine learning have significantly improved synthetic speech generation. This paper explores diffusion models, a novel method for creating realistic synthetic speech. We create a diffusion dataset using available tools and pretrained models. Additionally, this study assesses the quality of diffusion-generated deepfakes versus non-diffusion ones and their potential threat to current deepfake detection systems. Findings indicate that the detection of diffusion-based deepfakes is generally comparable to non-diffusion deepfakes, with some variability based on detector architecture. Re-vocoding with diffusion vocoders shows minimal impact, and the overall speech quality is comparable to non-diffusion methods.