Validation of forensic voice comparison by human analysts using the auditory-acoustic approach

The validation of methods used for forensic voice comparison (FVC) has received growing attention over recent years, largely driven by requirements of regulatory bodies and international standards. Many barriers have previously been identified by practitioners, including time and financial constraints and lack of existing resources. However, progress in establishing foundational validation in forensic voice comparison is vitally important; without it, the field risks falling behind the standards of other disciplines of forensic science and being viewed as unreliable in court. This study describes the largest and most comprehensive validation of the auditory-acoustic approach ever conducted and has been designed in consultation with a range of stakeholders to ensure its relevance and applicability. We conceptualise the task of auditory-acoustic FVC as a single, well-defined method employed by a competent analyst to arrive at a conclusion, which is then peer-reviewed by another competent analyst to generate an overall conclusion in the case. The study will involve a total of 80 speaker comparisons using forensically relevant recordings where the ground truth in relation to speakers’ identity is known. These involve a mixture of same-speaker and different-speaker comparisons and are divided equally between two experienced and qualified analysts. Each analyst will conduct primary analysis on half of the recordings and checking analysis on the other half. The aim of this validation exercise is principally to assess the extent to which the auditory-acoustic method in FVC is able to separate same-speaker and different-speaker pairs, evaluated using established metrics such as Equal Error Rate and minimum Log Likelihood Ratio Cost Function. In order to maximise transparency, the methodology for this validation exercise is being made public as a preprint on the Open Science Framework, prior to the comparisons being conducted by our analysts. The results will then be added to this document once complete.