Forensic Voice Comparison

Forensic Voice Comparison

Supervisors:
Yuko Kinoshita and Shun Ishihara

A recording of a human voice is a complex product of many factors. Some factors, such as recording channels or equipment, are external to the characteristics of the voice. Choices that speakers make consciously or subconsciously alter their voices too. Speech style, varieties of languages, and affective intentions: these all affect how we speak, but such effects are transient. Then, there are long-term factors which directly shape individual speakers’ characteristics, such as anatomy, linguistic background, and speech impediments. In forensic voice comparison, analysts have to tease out individual speaker characteristics from this complex mixture. In short, very challenging!

To tackle this complex challenge, we need to understand each component better. This means we have many useful research topics. For this summer project, we invite linguistics students with some phonetics background to explore one of three topics. You will use a novel acoustic feature — band-limited cepstral coefficients — as a voice comparison feature. The potential investigation topics are:

  1. Within- and between-speaker variability in consonantal segments, where it is difficult to make precise acoustic measurements. 
  2. Within- and between-speaker variability in different speaking styles.
  3. Within- and between-speaker variability in manifestation of emotions.

 

These projects require some skills in R programming language and Praat sound analysis software.

Updated:  1 August 2022/Responsible Officer:  Head of School/Page Contact:  CASS Marketing & Communications