Auxiliary objectives fail to improve speech spoofing detection reliably
Collapse, Not Invariance: Diagnosing Auxiliary Objectives in Speech Anti-Spoofing
Sound
Summary
Speech anti-spoofing systems try to detect fake or tampered voice signals, but their accuracy drops when the way the speech is generated or transmitted changes. The authors studied if adding extra goals (auxiliary objectives) to how these systems learn helps make them more reliable. After testing many options on different datasets and models, they found no consistent benefit over the standard method. Sometimes these extra goals even caused the system to collapse, producing almost constant outputs that hurt accuracy.
What this means in practice
- •For speech technology teams: Avoid relying on auxiliary objectives when improving anti-spoofing systems since they do not consistently enhance detection across models and conditions.
- •For cybersecurity engineers: Design voice authentication systems without auxiliary objectives for robustness because these add-ons can cause unpredictable failures in spoof detection.
Authors
Ksenia Lysikova, Kirill Borodin, Maxim Maslov, Grach Mkrtchian
Abstract
Speech anti-spoofing countermeasures degrade when the generator, codec or channel changes, and a common remedy is an auxiliary objective that shapes the embedding space; whether it does is invisible to EER, a pure ranking metric. We compare seven such objectives with cross-entropy over 113 runs on five corpora, AASIST3 at three seeds plus four pre-trained detectors, and measure the embedding space of the 24 AASIST3 runs directly. Raw augmentation displacement makes cosine consistency look effective, but the gain is a smaller space, not a more stable one: normalised by the spread, no configuration consistently improves on cross-entropy. Every trained space is dominated by the single decision axis expected for two classes, whose training-set structure does not transfer, and four runs collapse to a near-constant output that displacement rewards and EER reports as poor accuracy. No auxiliary objective keeps an advantage over cross-entropy across architectures, corpora and seeds.