Languages, dialects, and speakers can differ substantially in the temporal structure of speech. With the exception of only a handful of studies, the application of this information has been fairly limited in forensic speech research and casework. This may in part be due to existing differences in how to quantify temporal differences, as well as the limited research on the efficacy of such operationalizations for speaker discrimination. Standard operationalizations of temporal information have included measures reflecting global aspects of vowel or consonant duration alternations (e.g., Rhythm Metrics: RMs), as well as local measures of the change and acceleration of the cepstrum (e.g., delta and delta-delta coefficients in Automatic Speech or Speaker Recognition). This paper investigates the utility of these temporal measures for discriminating among four dialects of British English that contrast in region and language contact: Cambridge, Multicultural London, Leicester, and Punjabi-Leicester English. Using linear regression, log-likelihood model comparison and k-means clustering, we identified significant differences between dialects in all investigated RMs and substantially better performance of RMs in comparison to delta and delta-delta coefficients in dialect clustering. These findings suggest that temporal information in speech, and particularly global temporal information, is highly useful for dialect and speaker discrimination.
Cite as: Bradshaw, L., Hughes, V., Chodroff, E. (2020) Investigating the Forensic Applications of Global and Local Temporal Representations of Speech for Dialect Discrimination. Proc. Speech Prosody 2020, 635-639, doi: 10.21437/SpeechProsody.2020-130
@inproceedings{bradshaw20_speechprosody,
author={Leah Bradshaw and Vincent Hughes and Eleanor Chodroff},
title={{Investigating the Forensic Applications of Global and Local Temporal Representations of Speech for Dialect Discrimination}},
year=2020,
booktitle={Proc. Speech Prosody 2020},
pages={635--639},
doi={10.21437/SpeechProsody.2020-130}
}