Towards understanding the performance of individuals within automatic speaker recognition systems

Add to your list(s) Download to your calendar using vCal

• Vincent Hughes, University of York
• Monday 24 June 2024, 12:00-13:00
• Zoom only: https://cam-ac-uk.zoom.us/j/89860790207?pwd=sbErkKGhMwBJhyJcmlJ1jcQml5oTd4.1.

If you have a question about this talk, please contact Simon Webster McKnight.

Automatic speaker recognition (ASpR) systems are widely used in commercial settings for the purposes of personalisation and security, and are now increasingly being used to evaluate forensic evidence presented in court. Despite improvements in overall performance, even with forensically realistic materials (Morrison and Enzinger 2019), very little is known about why certain voices would be easy or difficult for systems to recognise.

In this talk, I will introduce a current ESRC -funded project (Person-specific ASR ) which attempts to better understand and explain individual variability in ASpR performance, with a particular focus on forensic applications. I will present data from a series of studies using both a small-scale, highly controlled dataset of recordings containing extreme vocal variation, as well as a large-scale, forensically realistic database provided to us by the UK Government. Finally, I will discuss our initial attempts to handle problematic speakers for ASpR systems, through condition adaptation and tailored calibration.

This talk is part of the CUED Speech Group Seminars series.

This talk is included in these lists:

• CUED Speech Group Seminars
• Cambridge Forum of Science and Humanities
• Cambridge Language Sciences
• Cambridge talks
• Chris Davis' list
• Guy Emerson's list
• Information Engineering Division seminar list
• PhD related
• Zoom only: https://cam-ac-uk.zoom.us/j/89860790207?pwd=sbErkKGhMwBJhyJcmlJ1jcQml5oTd4.1

Note that ex-directory lists are not shown.