• Christie Anne Jeannine Laurent
4. Term, Sound and Musik (Master Programme)
This document deals with voice synthesis techniques based on a combination of physical modelling and analytical elements, and serving as support for the design of an ageing voice model. The side contribution of this project is an adapted and complementary multiple-fold analysis tool that was developed in parallel. The aim of the whole project is to create a parametric ageing voice model where age becomes a tunable parameter, and this paper is meant to be its theoretical ground. Stated as above, this project digs into a barely explored branch of the voice synthesis field, even though voice synthesis is currently on-trend: numerous applications indeed exist nowadays, but very few consider age as a tunable parameter [Story et al., 2018, Schotz, 2006]. In a closely related branch, several have been trying to manipulate an existing voice to mak eit sound older or younger [Skoog Waller and Eriksson, 2016, Rupal and Seth, 2017]. For both examples, a certain knowledge about voice ageing is necessary; unfortunately, this is a complex phenomenon not yet fully understood at the current time. This document therefore gathers knowledge, theories and applications about the voice: it presents its production process and its characterisation, including its evolution over a lifetime; it addresses the physics and the physiology necessary to explain the previous elements and the different computing techniques employed to model it before introducing the ageing model that was developed. The fixed-age voice model (FAM) and the ageing voice model are finally evaluated in terms of credibility and quality.
Publication date5 Aug 2020
ID: 338343093