Countermeasures to Replay Attacks: A Review

Abstract

ABSTRACT Replay attack is an attempt of using pre-recorded speech samples of any target for acquiring unauthorized access to the automatic speaker verification (ASV) system. It is a low technology spoofing attack, requires only a high-quality recording and playback device, therefore, is the most accessible and highly effective approach of spoofing state-of-the-art ASV systems. Of late, the researchers are giving wide attention towards the development of replay attack countermeasures. This study provides a detailed review on the recently proposed replay attack detection methods. Different speech signal attributes, such as spectral magnitude features, modulation features, phase and excitation source features, have been explored for replay detection task. It is observed that the many proposed methods are performing well, but in light of day-by-day advancement in the device manufacturing technologies, even more effort is required towards the development of generalized replay attack countermeasures. In this study, we inferred that the exploration of excitation source information by suitable signal processing algorithms may be useful for replay detection task. They can also be used as complementary to spectral features to obtain generalized solutions. The LP residual signal represents excitation source information implicitly. As an immediate future scope, the potential of the LP residual signal information can be explored for the detection of replay signals. Alternatively, explicit excitation source information like pitch, epochstrength, and glottal flow derivative (GFD) signal can also be used for the detection of replay signals. This study is completed with the discussion on the frameworks for the proposed excitation information-based research directions.