On the contribution of the voice texture for speech spoofing detection

Automatic speech verification (ASV) consists in the implementation of automatic algorithm to measure and asses human biometric parameters. Serious vulnerabilities are emphasized concerning spoofing attacks. In this paper, we propose to investigate the possible contributions of voice textures to detect spoofing attack. Voice texture is a recent concept of voices characterization using an overall sound homogeneity. Well known spoofing attacks are based on text to speech (TTS), voice conversion (VC) and replay techniques. According to the nature itself of the spoofed voices, their textures are different from those of genuine speeches. The concept of the texture is well developed in the context of image processing. Local binary patterns (LBP) is one of the famous visual descriptor of the images. LBP was adapted to be used as speech texture descriptor. LBP coding is applied to all input genuine and spoofed signals. after that, histogram of the LBP descriptors is constructed and is used as features. Support Vector Machines (SVM) classifier is used to classify the obtained features as genuine or spoofed. From the experimental results, it is observed that the proposed method could increase the difference between genuine and spoofed speech.