Discriminatory Capacity of the Most Representative Phonemes in Spanish: An Evaluation for Forensic Voice Comparison

In this paper, a study of the discriminatory capacity of the most representative segments for forensic speaker comparison in Mexican Spanish is presented. The study is based on two corpora in order to assess the discriminatory capacity of the fundamental frequency and the three first vocalic formants acoustic parameters for reading and semi-spontaneous speech. We found that the context /sa/ has 73% of discriminatory capacity to classify speakers using the three first formants of the vowel /a/ with a dynamic analysis. We used several statistical techniques and found that the best methodology for the recognition of patterns consists of using linear regression with a quadratic fitting to reduce the number of predictors to a manageable level and apply discriminant analysis on the reduced set. This result is consistent with previous research data despite the methodology for Mexican Spanish had never been used.