ARISE will develop systems that use speech technology to handle automatically relatively simple telephone-based inquiries for rail travel schedule information. Other projects have shown that we now have the technology to build such systems but the main challenge is achieving a sufficiently high level of quality, particularly in areas where customer expectations of service levels are high. ARISE will install systems to measure caller satisfaction and assess the effect of improvements introduced by the project. Systems will be installed in Dutch, English, French and Italian customer environments. Initially the tests will be tightly controlled. As the project progresses, test coverage will be widened in each customer environment, building on successful outcomes of previous test phases. It is expected that the systems will be offered to the public at large at the end of the project.
ARISE is developing an automated enquiry service for railway timetables which will be accessed via an ordinary telephone and which will handle the bulk of routine enquiries automatically. The enquiry service will build on the successful results of two previous EU funded projects which used the emerging technologies of spoken language understanding automatically handling dialogues with spontaneous speech. One system will be produced for each of the Dutch and Italian railway operators, and two for the French railway operator. They will begin with a limited service offering timetable information which will be expanded later adding such features as ticket prices and 'door to door' information.
Schedule information is essential for users of public transport services which run at intervals greater than ten minutes. Usually it is available through different media. However, telephone enquiry systems play a crucial role, taking more than 200 million calls annually to railway centres in Europe, for example.
The throughput of a human operated service limits the number of calls that can be handled. Experience indicates that at least 20% of calls are unanswered. The delivery of the service may be of a high quality for the calls that are answered, but costs are high. Many of the calls answered are relatively simple to process.
ARISE is a response to this situation, and to other pressures to reduce costs. The railway operators in the project want to use speech technology to provide a solution to the problem by developing an automated service which is available by ordinary telephone. Other related projects have already demonstrated that the technology to build such systems is available, but the main challenge is achieving a sufficiently high level of quality, to service a market in which customer expectations are high.
The goal of the system is to handle the majority of routine enquiries automatically and maintain service quality, measured by caller satisfaction metrics. Such a system will allow operators to concentrate on more complex and higher added value services.
ARISE is organised around three language environments, each bringing together a user, technology providers and system integrators.
Dutch: the railway operator in the Netherlands handles mainly commuter traffic in two relatively short, peak periods of the day. To make the best use of investment and other resources it is important to increase off-peak traffic and one of the factors in achieving this is to provide timely information to the occasional traveller. A subsidiary company handles more than 9 million calls annually with 400 operators. Sixty percent of these calls require information concerning more than one public transport service.
One of the technology providers pioneered speech understanding technology and now has 20 years experience. They have achieved an 80% success rate for a train timetable enquiry system in Germany which has been operational since February 1994. One of the two university partners will supply the phonemic forms of names, the other its expertise in speech and pattern recognition. The system integrator has wide experience in building speech recognition systems.
In the Netherlands there are still technical and operational issues to resolve so that the existing system is suitable for wider deployment. These include finding ways of:
- extending the service to encompass multi-modal travel information
- integrating the automated system with human operator services to provide a full and flexible information service.
- providing information for 'door to door' journeys.
- improving performance in an existing prototype so as to reduce the number of misunderstandings about departure point, destination, time, day, etc.
- make dialogues flexible enough to handle customer enquiries more effectively, e.g. by facilitating the choice of later or earlier connections.
French: the railway operator in France receives more than 40 million telephone enquiries each year, and employs a staff of 1300 to answer them. It has extensive experience with public information systems over the telephone network, especially dialogue systems.
One of the technology specialises in speech analysis and synthesis. The university partner has experience in automatic speech processing, mainly in phonology, lexicons, and analysis of speech corpora. The two system integrators have particular experience in speech recognition and synthesis, and public transport information systems.
There are two existing systems which need to be improved in terms of both speech recognition and dialogue management. The improved system will be developed into a prototype service and the potential for a multi-lingual service investigated. The prototype will be used to:
- measure the quality of service, as perceived by the callers
- improve speech recognition performance and dialogue capability (involving issues of, accuracy, robustness and ergonomics and vocabulary size)
- add features to make reservations and give dynamic real time information
Italian: the railway operator in Italy deals with over 20 million telephone enquiries annually.
One of the technology providers has focused research on speech recognition in 'noisy' environments. The system integrator has already supplied a videotext train timetable and has considerable experience in on-line services.
The existing systems were developed as a result of the successful and RAILTEL pilot projects funded as part of the EC Multi-lingual Action Plan (MLAP) programme. In these projects, the technology demonstrated its potential for train timetable applications. The current prototype system enables telephone based, natural language enquiries, with user friendly interfaces in five languages.
The existing systems will be improved with new dialogue management strategy, continuous speech recognition and more flexible information reporting. The system will also be used in a set of field trials involving users from the public at large, in order to identify and analyse problems associated with large scale deployment.
One Dutch, one Italian and two French demonstrators will use national train schedule information to test their effectiveness. User test coverage will be systematically widened as the project progresses, and the degree of control over the tests will be gradually reduced.
The user test groupings are:
- Controlled: 50-100 participants. Participants are carefully selected and use of the system is closely monitored. Tests will follow fixed scenario's such as: 'you are in Paris and should attend a wedding in Lyon tomorrow at 14:00'.
- Expert: People familiar with the systems and the underlying issues and problems.
- Friendly: 500+ participants. A telephone number will be made available to a wider base of people within and outside the consortium and other 'friendly' contacts. Users will be able to call the number at any time to test the system.
- Public: Several thousand participants. A telephone number will be made available to the public so that they can call the system in the same way as a manually operated service. Users will consider the system to be operational. This trial will identify any remaining problems.
Funding SchemeCSC - Cost-sharing contracts
91198 Gif -Sur-yvette
2260 AK Leidschendam