Forensic Speaker Identification: a tutorial

This paper presents an overview of the methodologies used in the forensic field for the identification of speakers. First, an introduction shows the interest of the forensic community in speaker recognition and its limitations. Then the history of the evolution of speaker recognition systems -from Bell Laboratories in the fifties to current i-vector/PLDA approach-, and the much older history of forensic speaker identification that dates back to the seventeenth century at the court of Charles I of England -from recognition techniques by listening to semi-automatic systems nowadays-, are presented. To clarify concepts, some important concepts as speaker recognition tasks: verification and identification, and speaker recognition modalities: text-dependent/independent, are defined. Topics covered in this work range from particularities of forensic speaker identification tasks and complexities of collecting and processing speech samples, to final presentation of results to judge. Finally, a classification of forensic speaker identification approaches: auditory-spectrographic, auditory-perceptual, phonetic-acoustic/linguistic, automatic, semi-automatic and combined, are analyzed in depth. An international vision of the institutions involved in this forensic area is completed with a detailed analysis of what happens in Argentina. The final conclusion suggests the implementation of the combined approach, standardization of processes and quality certification of the forensic laboratories.