Dr. Joao Paulo Cabral

Summary

Research project to develop expressive Text-to-Speech commercial applications for industry. The project aims to validate prototype product/service and commercial value to companies that need AI expressive voice solutions.

Funding Agency

Enterprise Ireland

Programme

Commercialisation Fund

Summary

Through the growth of intelligent personal assistants, pervasive and wearable computing and robot based technologies speech interfaces are set to become a common dialogue partner. Technological challenges around the production of natural synthetic voices have been widely researched. Yet comparatively little is understood about how synthesis affects user experience, in particular how design decisions around naturalness (e.g. accent used and expressivity) impact the assumptions we make about speech interfaces as communicative actors (i.e. our partner models). Our ground-breaking project examines the psychological consequences of synthesis design decisions on the relationship between humans and speech technology. It fuses knowledge, concepts and methods from psycholinguistics, experimental psychology, and human-computer interaction (e.g. perspective taking and partner modelling research in human-human dialogue, controlled experiments, questionnaires) and speech technology (generation of natural speech synthesis) to 1) understand how synthesis design choices, specifically accent and expressivity, impact a user's partner model, 2) how these choices interact with context and 3) impact language production.

Funding Agency

Irish Research Council

Summary

The issue raised in this project is that although we know how to intuitively distinguish between speech and singing, there are portions of each in which one perceives the coexistence of both and this suggests that there is a gradation, more than an abrupt change in phonation and other aspects of speech. The aim of this research project is to focus on some aspects of production and perception of speech and singing in order to answer the question: Are speech and singing completely different phenomena? Experimental studies are conducted that include: collection of a corpus of spoken and singing data, measurements of acoustic differences between the two types of data, a perception test which aims to designate the presented stimulus as speaking or singing, and to use machine learning to further study the acoustic differences between the two sound categories. The results are analysed by taking into account cognitive aspects of speech and song.

Funding Agency

São Paulo Research Foundation (FAPESP)