DiffSSD: A Diffusion-Based Dataset For Speech Forensics

TL;DR

This paper introduces DiffSSD, a new dataset of 200 hours of synthetic speech from diffusion-based generators, to improve detection of such speech and address limitations of existing datasets.

Contribution

The paper presents DiffSSD, a comprehensive dataset for training and evaluating synthetic speech detectors on diffusion-based speech generation methods.

Findings

Existing detectors trained on older datasets perform poorly on DiffSSD.

DiffSSD improves detection accuracy for diffusion-based synthetic speech.

The dataset highlights the need for updated training data for speech forensics.

Abstract

Diffusion-based speech generators are ubiquitous. These methods can generate very high quality synthetic speech and several recent incidents report their malicious use. To counter such misuse, synthetic speech detectors have been developed. Many of these detectors are trained on datasets which do not include diffusion-based synthesizers. In this paper, we demonstrate that existing detectors trained on one such dataset, ASVspoof2019, do not perform well in detecting synthetic speech from recent diffusion-based synthesizers. We propose the Diffusion-Based Synthetic Speech Dataset (DiffSSD), a dataset consisting of about 200 hours of labeled speech, including synthetic speech generated by 8 diffusion-based open-source and 2 commercial generators. We also examine the performance of existing synthetic speech detectors on DiffSSD in both closed-set and open-set scenarios. The results highlight the importance of this dataset in detecting synthetic speech generated from recent open-source and commercial speech generators.