Combining Automatic and Phonetic-Acoustic Speaker Recognition Techniques for Forensic Applications

The central argument of the present study is that the performance of modern forensic speaker recognition can be improved by combining the strongholds of two entirely different approaches, that is: the classical acousticphonetic method and a potent automatic total-voice system such as IdentiVox. Since both make use of different and statistically independent sets of features of the phonetic and linguistic signal their individual results may eventually be combined into one conclusion regarding the key question of identity or non-identity between anonymous and suspected speakers. Two real-world cases will be presented as practical examples.