Speaker Line-up Calibration of the i-vector Based Speaker Recognition System for Forensic Application

An automatic speaker recognition (ASR) system must produce reliable likelihood ratios (LR) in order to be used for evaluating and presenting speech evidence to court. The LR is only reliable if it produced from a well-calibrated ASR. A study by Rodriguez (2007) showed that the LR calculated from the un-calibrated system was often misleading, while the calibrated system produced more reliable LRs. These results illustrate the necessity of performing LR calibration when using the ASR system for forensic applications. Several calibration techniques for ASR systems, such as the linear calibration method as presented in (Brümmer, 2006) and the novel line-up calibration method which was developed by van Leeuwen and Brümmer (2011) have been proposed in order to calibrate the LR produced by ASR.