Generalizing Voice Presentation Attack Detection to Unseen Synthetic Attacks and Channel Variation

Abstract

AbstractAutomatic Speaker Verification (ASV) systems aim to verify a speaker’s claimed identity through voice. However, voice can be easily forged with replay, text-to-speech (TTS), and voice conversion (VC) techniques, which may compromise ASV systems. Voice presentation attack detection (PAD) is developed to improve the reliability of speaker verification systems against such spoofing attacks. One main issue of voice PAD systems is its generalization ability to unseen synthetic attacks, i.e., synthesis methods that are not seen during training of the presentation attack detection models. We propose one-class learning, where the model compacts the distribution of learned representations of bona fide speech while pushing away spoofing attacks to improve the results. Another issue is the robustness to variations of acoustic and telecommunication channels. To alleviate this issue, we propose channel-robust training strategies, including data augmentation, multi-task learning, and adversarial learning. In this chapter, we analyze the two issues within the scope of synthetic attacks, i.e., TTS and VC, and demonstrate the effectiveness of our proposed methods.