This page is intended to illustrate the degree of auditory similarity among some of the speakers in the Young Australian Female Speakers Map-task (YAFM) database, which was collected between 2007 and 2009 to provide more challenging examples for forensic voice comparison. It has so far been used to investigate the strength of glottal pulse information in FVC (Vandyke et al. 2013); the nature of disfluencies in Australian and British English speech (Mc.Dougall et al. 2022); and for comparing formant trajectory and point measurements in FVC (Rose 2015). An annoymised version is freely available from Phil Rose for research purposes.
The YAFM database contains recordings of 26 (nearly all) young Australian females recruited from the close friends of the research assistant who devised and ran the experiment: from school, university, work and her soccer team. Many of them knew and interacted with each other. It was hoped that the choice of speakers homogeneous with respect to age, socio-economic background and close social group would tend to decrease the between-speaker variation and thus make any forensic voice comparison task more difficult. In addition, the elicitation sessions were all run by the research assistant herself, and it was expected that the enhanced interactivity between her and her mates would result in some typical in-group convergence, again contributing to a reduction in between-speaker variance. As is essential for FVC testing, non-contemporaneous recordings were made: 21 of the speakers were recorded twice, separated by about a week.
In the table below are ten comparisons between two recordings of voices from the database saying turn left at Earthworks road or something very similar. Try testing yourself to see how well you can identify whether two recordings are from the same or different speakers! It's not easy: there is only a little speech, the speakers are similar, and you are not familiar with their voices. The best approach is first to listen to all the recordings: this gives you an idea of the variation between them. Then go thru all ten comparisons, listen to the pair of recordings, and for each comparison write down whether you think it is the same speaker speaking, or different speakers.
You have no prior information as to how many same-speaker comparisons there are or how many different-speaker comparisons: they may all be same-speaker comparisons; or all different-speaker comparisons; or any mix in-between! So you have to judge each comparison on its merits - how it sounds to you.
Once you have gone thru all 10 comparisons, and made your decisions, scroll to the bottom of the page to find out how well you did!
|
comparison 1 |
||
|
comparison 2 |
||
|
comparison 3 |
||
|
comparison 4 |
||
|
comparison 5 |
||
|
comparison 6 |
||
|
comparison 7 |
||
|
comparison 8 |
||
|
comparison 9 |
||
|
comparison 10 |
References
Mc.Dougall, K. Duckworth, M. and Li, K. K. (2022) Individual and group variation in disfluency behavior in Australian English: a comparison with Standard Southern British English and York Englishes, SocioPhonAus Workshop.
Rose, P. (2015) Forensic Voice Comparison with Monophthongal Formant Trajectories - a likelihood ratio-based discrimination of "Schwa" vowel acoustics in a close social group of young Australian females, Proc. Int'l Conf. on Acoustics Speech & Signal Processing, Brisbane: 4819--4823.
Vandyke D., Wagner, M. and Rose, P. (2013) The Voice Source in Forensic-Voice-Comparison: a Likelihood-Ratio based Investigation with the Challenging YAFM Database, Conf. International Association of Forensic Phonetics and Acoustics.
Click here to open window with answer