Multi-Target Backdoor Attacks Against Speaker Recognition

TL;DR

This paper introduces a multi-target backdoor attack on speaker recognition systems using position-independent clicking sounds, achieving high success rates across multiple speakers and under varied noise conditions.

Contribution

It presents a novel multi-target backdoor attack method for speaker recognition, extending to verification tasks and analyzing trade-offs between stealth and effectiveness.

Findings

Achieves up to 95.04% success rate on multiple speakers.

Effective in speaker verification with success rates up to 90%.

Demonstrates trade-off between attack stealth and success.

Abstract

In this work, we propose a multi-target backdoor attack against speaker identification using position-independent clicking sounds as triggers. Unlike previous single-target approaches, our method targets up to 50 speakers simultaneously, achieving success rates of up to 95.04%. To simulate more realistic attack conditions, we vary the signal-to-noise ratio between speech and trigger, demonstrating a trade-off between stealth and effectiveness. We further extend the attack to the speaker verification task by selecting the most similar training speaker - based on cosine similarity - as a proxy target. The attack is most effective when target and enrolled speaker pairs are highly similar, reaching success rates of up to 90% in such cases.