A new speaker verification spoofing countermeasure based on local binary patterns

This paper presents a new countermeasure for the protection of automatic speaker verification systems from spoofed, converted voice signals. The new countermeasure is based on the analysis of a sequence of acoustic feature vectors using Local Binary Patterns (LBPs). Compared to existing approaches the new countermeasure is less reliant on prior knowledge and affords robust protection from not only voice conversion, for which it is optimised, but also spoofing attacks from speech synthesis and artificial signals, all of which otherwise provoke significant increases in false acceptance. The work highlights the difficulty in detecting converted voice and also discusses the need for formal evaluations to develop new countermeasures which are less reliant on prior knowledge and thus more reflective of practical use cases.