Impact of Frame Loss Aspects of Mobile Phone Networks on Forensic Voice Comparison

Abstract

The analysis of mobile phone speech recordings can play an important role in criminal trials. However it may be erroneously assumed that all mobile phone technologies, such as the Global System for Mobile Communications (GSM) and Code Division Multiple Access (CDMA), are similar in their potential impact on the speech signal. In fact these technologies differ significantly in their design and internal operation. This study investigates the impact of an important aspect of these networks, namely Frame Loss (FL), on the results of a forensic voice comparison undertaken using a Bayesian likelihood ratio framework. For both networks, whenever a frame is lost or irrecoverably corrupted, it is synthetically replaced at the receiving end using a history of past good speech frames. Sophisticated mechanisms have been put in place to minimize any resulting artefacts in the recovered speech. In terms of accuracy, FL with GSMcoded speech is shown to worsen same-speaker comparisons, but improve different-speaker comparisons. In terms of precision, FL negatively impacts both sets of comparisons. With CDMA-coded speech, FL is shown to negatively impact the accuracy of both same- and different-speaker comparisons. However, surprisingly, FL is shown to improve the precision of both sets.