Ulrich Scherhag successfully defended his Master thesis „Presentation Attack Detection for State-of-the-Art Speaker Recognition Systems“, after working with da/sec in the period of October 2015 to April 2016.
In the modern society, biometrics is gaining more and more in impor- tance, particularly facilitated by the increasing usage of multi-factor authentication. Due to the advancing distribution of mobile phones, speaker recognition plays a special role.
Despite all the advantages of biometrics, the vulnerability of the sys- tems against attacks is still an existing weakness. In particular, speaker recognition systems are threatened. Due to the sophisticated research in the field of speech synthesis, a wide rage of effective methodolo- gies for attacking speaker recognition systems is easy to utilize: i.e. replay, speech synthesis and unit-selection.
State-of-the-art countermeasures are successful in detecting synthe- sis and voice conversion attacks, but fail on detecting unit-selection attacks. The impact of these attacks to state-of-the-art speaker recog- nition systems is analysed. Thus, the focus of this thesis motivates: the creation and detection of unit-selection attacks, proposing a new countermeasure based on the principle of frequency analysis. For the evaluation of the experiments the metrics introduced in ISO/IEC CD2 30107-3 are utilized. Detection techniques of current research are dis- cussed and new detection algorithms proposed. In contrast to con- ventional attack detection algorithms, which utilize feature extraction methods, known in the field of speech recognition for modelling the perception of sound by the human ear, the proposed algorithms dis- claim the use of these filters and analyses the unfiltered frequency band.