Vocalis in EC project to create the first industry standards for spoken dialogue systems

A European Commission (EC) research project, DISC, is on the way to establishing the first industry standards for one of the fastest growing and most challenging areas of information technology – speech recognition-based user interfaces.

The project involves commercial and academic partners across Europe. They include Cambridge-based Vocalis, a world leader in speech recognition technology, together with major industrial and research organisations such as Daimler-Benz and France's National Centre for Scientific Research.

Speech recognition is a rapidly emerging technology that has huge potential for creating easy to use, voice-driven interfaces for many kinds of applications, in such areas as call centres and telecommunications.

Clearly, for such interfaces to work, the recognition must be excellent. But there is more to designing effective spoken language dialogue systems (SLDSs) than sheer recognition accuracy. Other features, such as the structure of the dialogue, how effectively it works in practice, how well it matches the needs and abilities of its target users, and how best to integrate it with other systems, are equally important. All these are addressed in the DISC project.

"At this time there are no accepted standards or even widely understood benchmarks for assuring potential customers of SLDSs of the quality of systems, or for comparing SLDSs before deployment in the field," says Klaus Failenschmid, a Dialogue Researcher for Vocalis. "This situation continues to generate uncertainty about SLDSs, in terms of their application, usability and cost."

"The aim of DISC is to contribute towards establishing dialogue engineering as a sub-discipline of software engineering through developing a detailed and integrated set of development and evaluation methods and procedures. This will create the first best-practice model for dialogue engineering, as well as a range of support concepts and software tools."

DISC began by assessing a range of SLDS products designed and implemented over the last decade. These included:

• The Vocalis SPEECHtel Voice Activated Dialling application and Operetta, a voice activated auto attendant
• A dialogue manager and natural language parser from Daimler-Benz
• A Danish system for flight ticket reservation from Odense University, Denmark
• The Vermobil translation system - a long-term project of the German Federal Ministry of Education, Science, Research and Technology (BMBF) with 29 industrial and academic partners.
• The Waxholm tourist boat information system from KTH (Kungliga Tekniska Hoegskolan), Sweden, an academic institution in Stockholm

From these studies DISC developed its two central components, a ‘grid’, which contains a series of checklist entries, and a life-cycle model, which aims to capture the development and evaluation process of SLDSs.

The project has now entered its second phase, which will involve inviting colleagues from outside the DISC participants to comment on and apply the grid and life cycle. From this, best practices and standards for SLDSs will emerge. A proposal has already been made for a further project, DISC2, which will make this information more widely available and provide practical tools for SLDS development.

The participants in DISC are:
MIP, Odense University, Odense, Denmark
KTH, Department of Speech, Music and Hearing, Stockholm, Sweden
LIMSI/CNRS, Human-Machine Communication Department, Orsay Cedex, France
Vocalis Ltd, Cambridge, UK
IMS, Stuttgart University, Germany
Daimler-Benz, Ulm, Germany
Linkping University, Department of Computer and Information Science, Linkping, Sweden.

The DISC project has its own web page at: http://www.elsnet.org/disc/