Spoofing-Aware Speaker Verification via Wavelet Prompt Tuning and Multi-Model Ensembles

TL;DR

The paper addresses the unified task of Spoofing-Aware Speaker Verification (SASV) under the WildSpoof 2026 challenge, where systems must simultaneously verify identity and authenticity. It introduces a cascaded approach that combines a Wavelet Prompt-Tuned XLSR countermeasure with an AASIST-backed spoof detector and a three-model ASV ensemble fused via Z-score methods. The CM achieves a parameter-efficient design with $1\times 10^6$ trainable parameters, while the ASV ensemble leverages ResNet34, ResNet293, and WavLM-ECAPA-TDNN to deliver robust speaker verification; the integrated system attains a Macro a-DCF of $0.2017$ and a SASV EER of $2.08\%$, with in-domain spoof detection as low as $0.16\%$ EER. However, cross-domain generalization to unseen attacks (e.g., ASVspoofF5 and ASV 2022) remains challenging, underscoring the need for more diverse training data and attack scenarios to bolster SASV robustness in real-world deployments.

Abstract

This paper describes the UZH-CL system submitted to the SASV section of the WildSpoof 2026 challenge. The challenge focuses on the integrated defense against generative spoofing attacks by requiring the simultaneous verification of speaker identity and audio authenticity. We proposed a cascaded Spoofing-Aware Speaker Verification framework that integrates a Wavelet Prompt-Tuned XLSR-AASIST countermeasure with a multi-model ensemble. The ASV component utilizes the ResNet34, ResNet293, and WavLM-ECAPA-TDNN architectures, with Z-score normalization followed by score averaging. Trained on VoxCeleb2 and SpoofCeleb, the system obtained a Macro a-DCF of 0.2017 and a SASV EER of 2.08%. While the system achieved a 0.16% EER in spoof detection on the in-domain data, results on unseen datasets, such as the ASVspoof5, highlight the critical challenge of cross-domain generalization.