Speech Spoofing Detection In Voice Based Biometric System

Tracking multiple passwords,PINs,memorable dates,and other authentication details requires remote access to one of the less attractive challenges of life.Using voice-based verification as a biometric technology for both children and adults can be a good alternative to old memory-based procedure.Using voice for authentication may be useful in many application domains,including security,protection,education,communication-based services,and Webbased services.However,voice-based biometrics applications are subject to different types of spoofing attacks.The easiest and affordable subversion for a voice spoofing attack is a replay attack.With a pre-recorded speech sample,replay attack offers a real risk for automatic speaker verification systems.Many countermeasures have been developed recently.Most efforts focus on finding more distinctive features and many new features have been suggested.Five types of features namely Mel Frequency Cepstral Coefficients(MFCC),Linear Frequency Cepstral Coefficients(LFCCs),inverted Mel Frequency Cepstral coefficients(IMFCCs),Constant Q Cepstral Coefficients(CQCCs),and bottleneck feature were analyzed with the public data sets ASV spoof 2017 and BTAS 2016.The experimental results show that MFCCs and bottleneck features give similar results.Both of them significantly outperform others(including recently proposed CQCCs).However,the number of filters and Cepstral bins are essential to the success of MFCCs.Furthermore we used recurrent neural networks for automatic replay spoofing attack detection,we focused on recurrent neural networks with better recurrent units that involve gating mechanism,such as a Long Short Term Memory(LSTM)unit and Gated Recurrent Unit(GRU).Our results shows that neural networks performed better then GMM on ASV Spoof 2017 and on BTAS 2016 we got best results of GRU models.