SynSFX: Multi-Model Sound Effects Synthesis Dataset for Deepfake Detection and Evaluation

2026-07-06

AI Summary

This work addresses the limited generalization of existing audio deepfake detection methods on synthetic sound effects and the absence of large-scale, multi-source benchmark datasets. To bridge this gap, we introduce SynSFX, a novel dataset comprising 43,374 annotated audio clips—26,452 synthetic and 16,922 real—systematically generated using seven state-of-the-art text-to-audio models. SynSFX is the first large-scale resource to offer diverse synthetic environmental sound effects spanning multiple generative architectures, thereby filling a critical data void in sound-effect-oriented deepfake research. By providing comprehensive coverage of modern synthesis techniques, the dataset substantially enhances the evaluation foundation and generalization capabilities of deepfake detection approaches.

Abstract

While audio deepfake detection has advanced significantly, representative detectors show limited generalization to synthetic sound effects. Existing environmental audio datasets such as EnvSDD provide important initial resources, but remain limited in scale and generation provenance for studying isolated sound-effect deepfakes. To support this direction, we present SynSFX, a large-scale corpus of 43374 clips (26452 synthetic, 16922 real) spanning 7 popular text-to-audio models.

Problem

• audio deepfake
• sound effects
• synthetic audio
• deepfake detection
• dataset

Innovation

• sound effects synthesis
• audio deepfake detection
• text-to-audio generation
• SynSFX dataset
• multi-model evaluation