Personification Using Affective Speech Synthesis: An Introduction

Matthew Aylett

University of Edinbugh, Informatics
Tuesday, September 18, 2012
12:30 PM, Conference Room 6A
 

Abstract

For many applications, multimodal systems require the ability to convey personality. This requirement varies from the explicit, where a virtual agent is mimicking a human in an immersive training application[1], to the implicit, effective interaction in many systems that is improved due to the enhanced involvement and pleasure of using a multimodal system which creates a sense of a personality[2].

If this multimodal system is required to communicate with humans using audio, then speech synthesis is a critical element in rendering the sense of personality. In this paper we discuss the issues raised by the difficulty of assessing personality in artificial systems as well as the possible strategies for enhancing speech synthesis in order to create a deeper sense of personality rather than the more traditional objective of naturalness.

We examine work in expressive speech synthesis and present work carried out at CereProc in order to create emotional synthesis. The expressive functionality of CereVoice[3] is available in a downloadable application and has been used in order to produce some novel character-based android voices. We will examine the elements of this system in terms of accent, voice adaptation, and expressive synthesis.

We will then present preliminary results from an emotional assessment of the CereProc system, which will form part of a baseline of evaluation for further assessment of personality. To conclude we will present our proposed personality assessment of this speech material.

The work presented forms part of the 'personify' project supported by the Royal Society, which commenced on January 1, 2012.

[1] L.Hall, S.Jones, A.Paiva, and R.S Aylett, “Fearnot!: providing children with strategies to cope with bullying,” in 8th International Conference on Interaction Design and Children, 2009.

[2] T.Bickmore, L.Pfeifer, and B.Jack, “Taking the time to care: Empowering low health literacy hospital patients with virtual nurse agents,” in SIGCHI Conference on Human Factors in Computing Systems, 2009.

[3] Aylett, M.P., Pidcock, C.P., “The CereVoice Characterful Speech Synthesiser SDK”, AISB, Newcastle. pp.174-8, 2007.

Speaker Biography

Dr. Matthew Aylett has more than 10 years experience in commercial speech synthesis and speech synthesis research. He is a founder of CereProc, which offers unique emotional and characterful synthesis solutions and has recently been awarded a Royal Society Industrial Fellowship to explore the role of speech synthesis in the perception of character in artificial agents.