Original Investigation of Forensic Voice Analysis Reliability for Speaker Identification in Nigerian Languages

Forensic voice analysis for speaker identification relies on acoustic and phonetic features whose discriminatory validity is dependent on the linguistic and phonological characteristics of the target language, creating significant reliability concerns when methods developed for European languages are applied to Nigerian linguistic contexts. This doctoral study makes an original contribution to forensic phonetics by investigating the reliability of forensic voice analysis methods for speaker identification in Hausa, Igbo, Yoruba, and Nigerian Pidgin English, the four most widely spoken languages and varieties in the Nigerian forensic evidence context. A multi-phase empirical design was employed, recording 200 speakers of each language variety under both controlled studio and degraded field conditions simulating Nigerian telephone intercept quality, extracting acoustic feature sets including fundamental frequency parameters, formant trajectories, voice quality measures, and language-specific phonological features, and applying likelihood ratio-based automatic speaker recognition and auditory-phonetic expert analysis to evaluate discriminatory accuracy and error rates. The study develops language-specific acoustic reference data and evaluates the performance of generic forensic speaker recognition systems against language-adapted models. Findings demonstrate that generic systems achieve false positive rates of eleven percent for Hausa and fourteen percent for Igbo under field conditions, compared to four percent for language-adapted models. An original Nigerian Forensic Voice Analysis Protocol is developed, incorporating language-specific feature weighting and degradation correction procedures. Recommendations include judicial guidelines for the admissibility of voice identification evidence in Nigerian courts. Keywords: forensic voice analysis, speaker identification, Nigerian languages, acoustic phonetics, likelihood ratio.