An Approach to Detect Replay Attack in Automatic Speaker Verification System

摘要

Automatic Speaker Verification (ASV) system is to accept or reject a claimed identity based on a speech sample. Spoofing, in general, is a fraudsters or malicious practice in which communication is sent from an unknown source disguised as a source known to the receiver. Spoofing is most prevalent in communication mechanisms that lack a high level of security in biometric authentication. It is proposed to develop an approach to detect replay attack in ASV system. The proposed system uses ASVspoof2017 database. In speech signal processing traditional cepstral processing is combined with constant Q transform. Spoofing countermeaasure has been made effective by the recent introduction of the resulting constant Q cepstral processing(CQCCs). The merits of CQCC features is due to a variable spectro-temporal resolution which also does reliable capture of indicative signs of spoofing attacks but basically varies from those that used by most ASV systems. The extracted CQCC feature are trained using Gaussian Mixture Model(GMM). Using the training utterances of genuine and spoofed speakers the respective model is created using GMM. The experiments shows that proposed system is significantly better using CQCC features.