Event summary: Acoustical Society of America conference (in Sydney)
The Hub had a presence at the Acoustical Society of America conference, which was held in Sydney from Dec 4-8, 2023.
I presented some work that Helen and I have been conducting in collaboration with Dr Kirsty McDougall, who is an Assistant Professor of Phonetics at University of Cambridge. Kirsty works on various topics in forensic phonetics, as you can see from her website. I gave a poster presentation called A preliminary investigation of the acoustic factors impacting decision-making in speaker attribution.
Speaker attribution is the task of assigning an utterance to an individual speaker. This can be a very high stakes exercise, because attributing an incriminating utterance to an individual aligns them with having committed a crime. This is an under-researched area, and we are aiming to bring more attention to the issues involved.
In our study, we looked at how five trained phoneticians attributed utterances to speakers in a short (97 second) recording. The audio was broadcast quality, the speakers were stationary and all had a microphone, but the material was challenging because there were multiple voices, overlapping speech and background noise. The results showed that some of the speakers were effectively indistinguishable because of shared acoustic features in their speech.
You can have a listen to the audio that our participants were presented with – have a think about how many speakers are talking, and which utterance belongs to which speaker. It is not easy!
You can access a copy of the Loakes Fraser and McDougall poster here, which explains how many speakers there were, and also talks about what our participants found most difficult in the process. People at the conference also expressed surprise at how hard this task was. I played the audio to some of the visitors at the poster, and people could not work out how many voices they were hearing (similar to the experience of our participants).