Exploring the Potential of Using Acoustic Patterns Extracted from Yeah in Forensic Voice Comparison

This exploratory study focuses on the discriminatory value of “higher order” features for forensic voice comparison (FVC). The phonetic variation of commonly used words could, in principle, provide good FVC features as they are relatively frequent in speech. Few studies have attempted to establish the validity of frequent words or word-like forms as speaker discriminants. Exceptions include filled pauses, yes , and just , each of which demonstrated good strength of evidence. However, frequent words are also used to achieve a variety of discourse functions, and there is some evidence suggesting that the pragmatic use of features can impact their pronunciation. In the present study, dynamic formant estimates from yeah were analysed in mobile phone speech data from 33 male speakers from West Yorkshire. Each token was tagged according to the broad pragmatic function of incipiency, allowing for observations both of speaker-specific and pragmatic-phonetic variation of yeah . The data were assessed via likelihood ratio–based testing, with yeah proving to be very poor at discriminating speakers (log likelihood ratio cost function >1). Speakers produced a flatter trajectory for yeah when it was in a freestanding turn, whereas in a full and longer turn, speakers produced yeah with more formant trajectory movement—this high level of phonetic variation across pragmatic categories indicates that yeah represents multiple functional units, making it a feature that may be too inconsistent for discriminating between speakers in FVC.