Synthetic Speech Source Tracing using Metric Learning

TL;DR

This paper explores source tracing in synthetic speech using metric learning and ResNet, demonstrating competitive results and highlighting the potential of speaker recognition techniques in audio forensics.

Contribution

It introduces a metric learning approach for source tracing in synthetic speech and evaluates its effectiveness against SSL-based methods using the MLAADv5 benchmark.

Findings

ResNet with metric learning matches or exceeds SSL-based systems.

ResNet is viable for source tracing in synthetic speech.

Optimizing SSL representations could improve results.

Abstract

This paper addresses source tracing in synthetic speech-identifying generative systems behind manipulated audio via speaker recognition-inspired pipelines. While prior work focuses on spoofing detection, source tracing lacks robust solutions. We evaluate two approaches: classification-based and metric-learning. We tested our methods on the MLAADv5 benchmark using ResNet and self-supervised learning (SSL) backbones. The results show that ResNet achieves competitive performance with the metric learning approach, matching and even exceeding SSL-based systems. Our work demonstrates ResNet's viability for source tracing while underscoring the need to optimize SSL representations for this task. Our work bridges speaker recognition methodologies with audio forensic challenges, offering new directions for combating synthetic media manipulation.