A CAUTIONARY TALE FOR PHONETIC ANALYSIS: THE VARIABILITY OF SPEECH BETWEEN AND WITHIN RECORDING SESSIONS
Sula Ross, Earnshaw, Kate, Erica Gold · Huddersfield Research Portal (University of Huddersfield) · 2019
This paper investigates within and between session variability using a subset of 60 British English male speakers from the WYRED project. Three separate speaking tasks were compared using extracted i- vector PLDA scores within iVOCALISE. Different speaker pairs from contemporaneous (within-session) recordings and non-contemporaneous (between- session) recordings were tested. A within-session, between-task comparison was also performed in order to consider variation in speech style in addition to non-contemporaneity. EER and Cllr values indicate that non-contemporaneity is not the only factor which needs to be taken into account when conducting phonetic analysis or evaluating speaker comparison systems, as speech style also seems to play an important role. Further analysis supports the requirement for (forensic/socio-) phoneticians to sample data from the entirety of a recording, especially if the nature of the speech elicitation may change during the task, as the degree of variability is dependent on which portion of the sound file is sampled.