Temporal Parameters of Spontaneous Speech in Forensic Speaker Identification in Case of Language Mismatch: Serbian as L1 and English as L2

The purpose of the research is to examine the possibility of forensic speaker identification if question and suspect sample are in different languages using temporal parameters (articulation rate, speaking rate, degree of hesitancy, percentage of pauses, average pause duration). The corpus includes 10 female native speakers of Serbian who are proficient in English. The parameters are tested using Bayesian likelihood ratio formula in 40 same-speaker and 360 different-speaker pairs, including estimation of error rates, equal error rates and Overall Likelihood Ratio. One-way ANOVA is performed to determine whether inter-speaker variability is higher than intra- speaker variability across languages. The most successful discriminant is degree of hesitancy with ER of 42.5%/28%, (EER: 33%), followed by average pause duration with ER 35%/45.56%, (EER: 40%). Although the research features a closed-set comparison, which is not very common in forensic reality, the results are still relevant for forensic phoneticians working on criminal cases or as expert witnesses. This study pioneers in forensically comparing Serbian and English as well as in forensically testing temporal parameters on bilingual speakers. Further research should focus on comparing two stress-timed or two syllable-timed languages to test whether they will be more comparable in terms of temporal aspects of speech.