Design and Testing of a Corpus for Forensic Speaker Recognition Using MFCC, GMM and MLE
Abel Herrera-Camacho, Adrián Zúñiga-Sainos, Gerardo Sierra, José Trangol-Curipe, Margarita Mota-Montoya, Adonay Jarquín-Casas · 2019
The importance of applying speaker recognition systems in forensic environments has increased in this century. One reason is the use of audio recordings as evidence in trials of every kind. This article presents a voice corpus design with this use in mind, based in recordings from speakers of the Spanish language dialect used in Central Mexico, distinct from any other corpus currently in use. For its evaluation, we used a Mel frequency Cepstral coefficient with a Gaussian Mixture Models for parametrization, and a Maximum Likelihood Estimation approach for classification. Results show an accuracy of more than 93% identification of the speaker at any condition, proving a good recognition model.