Mauro Nicolao works on the homeService project.
homeService is concerned with clinical applications of speech technology and is one of the two test applications for the NST programme of research.
The area of assistive technology is particularly well suited to demonstrate the impact of a project concerned with Natural Speech Technology.
For physically disabled and elderly users who either can't or choose not to use conventional methods of access to electronic devices such as remote controls,
PC keyboard-and-mouse, using speech can be a very natural and attractive alternative.
At the core of homeService is a longitudinal study of up to 10 elderly and physically disabled users of voice-enabled assistive technology in the domestic environment.
The main focus of the project is to further develop theoretical foundations for research into the recognition of disordered and elderly speech, looking at how best to harness the typical speech knowledge and data to achieve high performance for non-typical speech domains. The project also addresses research questions which may generalise to all forms of speech such as issues associated with systems evolving and adapting over time as more and more domain specic data becomes available.
Automatic speech recognition of disordered speech poses a number of challenges compared to recognition of typical speech. Disordered speech is generally far more variable and irregular than typical speech, and speakers typically also display a more limited vowel space as well as issues with timing and pitch control. Another challenge is the lack of suitable training data: the largest databases of disordered speech still contain orders of magnitude less data than what is used for state-of-the-art systems.
One of the outcomes of the homeService project is that, as the users start using the system, all the recorded audio will be collected and released back to the other researchers working on NST for further off-line research.
In the first year of the homeService project, focus has been on establishing a baseline recognition system for dysarthric speech and exploring various adaptation strategies. During the remaining years, users will be enrolled in the study. Each user's system will be independently maintained and will over the years become increasingly adapted to the individual user's needs both in terms of the modelling capabilities of the speech technology (such as acoustic and language models) but also in terms of functionality and types of services available. This will be accompanied with work on the theoretical side exploring the newly collected, unique data resource.