Accessibility statement

Music Coding Collective: Voice and Accent

Friday 12 November 2021, 11.30AM

The what, who and how of the human voice

Sarah Knight, Psychology

Human voices are richly informative, allowing the listener to draw conclusions about not just the speaker's message but also their identity, personality and affective state. Furthermore, listeners are able to do this even under challenging listening conditions, such as in a noisy restaurant or via a bad telephone line. Research into voice perception therefore needs to explore how we can understand not just what is being said, but also who is speaking and how they might behave or feel.

I will briefly outline some of the projects that I have been involved with in each of these areas, with particular emphasis on their relevance to voice-based technologies. I
will also highlight the need to consider these different facets of the voice simultaneously, as well as the importance of incorporating information from other modalities (e.g. face perception) into our understanding of vocal communication.

Anatomy, acoustics and the individual

Amelia Gully, Language and Linguistic Science

Given how variable even the same person's voice can be depending on their situation, health, mood, interlocutor, and any number of other factors, the task of identifying an individual from their voice is much more complicated than with other biometrics like fingerprints or DNA. However, there are some things that are (relatively) constant for a speaker, including size and shape of their vocal anatomy.

In this talk I will describe how we can use technologies such as magnetic resonance imaging (MRI) to collect information about speakers vocal tracts, and how we can use this data to inform detailed numerical acoustic simulations, allowing us to study the precise impact of changes in vocal anatomy on speech. I will present some data illustrating just how different the vocal tract can be between individuals and introduce some work in progress on the acoustic impact of such changes. I will also argue against the common practice in speech science of "averaging out" the idiosyncratic characteristics of individual speakers.

More information can be found on the Music Computing and Psychology Lab website.

Location: Online