Abstract: This Ph.D. Thesis proposes a complete methodology for the adaptation of automatic speaker recogition technology to forensic evaluation of the evidence. The nature of the problem of forensic automatic speaker recognition is deeply analyzed in the context of the current debate about scientific procedures in forensic disciplines worldwide. Then, a solution for this problem is proposed in the form of a hyerarchical methodology which integrates current standards and state of the art of automatic speaker recognition technology and the requirements and needs of the so-called coming paradigm shift in forensic science. The Thesis contributions are supported by numeorus peer-reviwed publications in national and international conferences and jorunals included in ISI-JCR. Also, this Thesis and its constributions have been the recipient of several awards in different national and international contexts. Moreover, the applicability of the Thesis is evidence by the multiple public and private contracts and projects which consider the framework presented here, as well as the impact of the proposed methodologies in important fora such as working groups of the European Network of Forensic Science Institutes.
Index Terms: Forensic speaker recognition, likelihood ratio, calibration, empirical cross-entropy, coming paradigm shift.