Abstract:

This article aims to investigate whether applying Lindley’s formula to speech data is justified. Our first problem is not knowing the true LR for any continuous parameters extracted from speech data: there have been no forensic speaker-identification studies with sufficiently large samples. However, with a good understanding of the parameter tested, we can to some degree predict how the LR should behave. How closely the LR estimates produced by Lindley’s formula match what is known about the parameter will tell us whether Lindley’s formula produces reliable LR estimates for speech data. The results of the experiments in this study seem to indicate that Lindley’s formula is indeed a useful tool to produce LR estimates of speech evidence, despite its omission of the occasion-to-occasion variation factor. The potential of long-term F0 as an additional piece of speech evidence seems to be very small, given that the LR of the test was found to be practically unity.

Author: Yuko Kinoshita