Michael Schillo's research page on

CORPUS-BASED INTERFACE TECHNOLOGY

The in-car personal assistant interface

Abstract

I researched the design of a Voice-User Interface sublanguage, a subset of English which allows users of a speech interface to issue commands in a reasonably 'natural' way while achieving acceptable speech recognition accuracy. The sublanguage should be tailored to cover the underlying functionality of the software system being interfaced to, but less constrained than a menu-keyword system: users should not be expected to have to learn a new language, but speak something near to "natural" English.

Thesis

My complete Thesis can be found here: A downloadable postscript.

Data

This is the data collected in the experiments.

In Detail

My research contributes towards a larger ongoing research and development programme, involving collaboration between the School of Computer Studies and Visionair, Ltd. Visionair are providing access to hardware and a practical application: the in-car "Intelligent Personal Assistant" system for travelling executives. The in-car application involves:

- a Psion hand-held computer "Personal Organiser", including diary, databases of addresses/appointments/etc

- a Global Positioning System (GPS) receiver unit, which can tell us (via satellite navigation) our exact position

- a Geographical Information System (GIS) to correlate to the information from the receiver to relevant locations in the Psion database

- a portable PC-based commercial Speech Recognition system and development environment (Speech Systems PE500); other speech recognition systems are also available for evaluation (eg CUED AbbotDemo)

Our long-term aim is: to add a computer system that can access the information in the Psion personal organiser (including access to connected GPS, GIS, and cellular telephone) via a Voice-User Interface (VUI), ie via a speech recognition system (such as PE500), so that it is possible to access any desired information via natural audio language input; furthermore, to extend this computer system beyond an interface, to include some of the "intelligence" and organisational functionality expected of a human Personal Assistant.

Visionair are developing the application: software systems on the Psion which the VUI will access. Visionair are already sponsoring a PhD student, Gavin Churcher. Mr Churcher's PhD programme is about "Integrating linguistic constraints in mobile speech systems"; his research is more general, including other applications (Air Traffic Control), and focusing on theoretical models of discourse, semantics and syntax to be integrated into speech recognisers.

My MSc research programme focused specifically on the design of an English-like sublanguage for the in-car Personal Assistant. My approach involved collation and analysis of a Corpus of expected typical spoken interactions. Ideally we wantedto allow an unconstrained range of possible utterances; but a fully-comprehensive corpus that would cover all possible spoken inputs (eg spoken dialogues from the British National Corpus) would force the PE500's performance to be very low. What we are trying to do is to find a model of the subset of natural language we are looking at with more constraints. This would simplify the syntax and increase the PE500s accuracy and overall performance. The fundamental base of such a model is a study of the possible utterances that occur.

Back to my research page

You can contact me on schillo@virtosphere.de