On robustness of speech based biometric systems against voice conversion attack

摘要: Graphical abstractDisplay Omitted HighlightsEvaluation of robustness SID and SV systems against VC spoofing attack.The vulnerability in decreasing order techniques is GMM, WFW WFW-.In systems, GMM-SVM more resilient than GMM-UBM for impostor attacks.All are robust to cross-gender intra-gender converted voices.An approach relating the score with performance proposed. Voice conversion (VC) approach, which morphs voice a source speaker be perceived as spoken by specified target speaker, can intentionally used deceive identification (SID) verification (SV) that use speech biometric. attacks imitate particular pose potential threat these kinds systems. In this paper, we first present an experimental study evaluate such disguise. We Gaussian mixture model (GMM) based GMM universal background (GMM-UBM) supervector support vector machine (GMM-SVM) this. conducted using three different techniques: technique, weighted frequency warping (WFW) method its variation, where energy correction disabled (WFW-). Evaluation done conversions between fifty male female speakers taken from TIMIT database. The result indicated degradation percentage correct (POC) equal error rate (EER) all Experimental results show most vulnerable towards WFW- conversion. From results, it also said that, general terms, slightly voices through This work extended find out relationship objective system CMU ARCTIC database, parallel corpus. experiment on quantifying related ability spoof system.