The Speechdat Project
Project Summary
Spoken Language Resources (SLR) are speech databases including associated
annotations, pronounciation lexica, and language modeling materials
needed for the development and use of speech recognition and
speech synthesis technology. Now and in the near future SLR are needed
on the one hand to develop applications using speech driven dialog systems
based on the available speech recognition technology and
on the other hand to develop the speech recognition technology
handling real spontaneous speech leading to products around the year 2000.
In the near term European Companies active in the area of speech driven
applications will have most success in the area of telecommunication,
because there exists a large European basis of telecom products. Teleservices,
which will be partly or fully automated using modern speech technology,
comprise a market of several billion ECU/year in Europe. There will be
a strong competition by US-companies and US-telecoms, which benefit from
their large "single language" economic basis and will profit from the
deregulation of the European telecom market.
Further the long term competitiveness of European companies must be secured
by supporting both public and private research in advanced speech technology.
SLR are costly to produce, because a lot of skill and manpower is needed
for each language. The goal of this project is to minimize the costs
by creating an infrastructure to ensure:
- efficient and fast production of SLR for short term and midterm
speech driven products
- efficient production of SLR for research in speech technology
- efficient distributtion of SLR taken in account IPR and
reusability issues
- definition and production of SLR for future research.
The project goals are achieved by:
- evaluation of the industrial market for short and midterm needs for SLR
- evaluation of long term needs of SLR
- proposing a model for data base description
- proposing a model for working standards ,distribution and production
- making a case study, which is based on a concrete model for database
production, working standards and distribution.
The infrastructure, defined by the models, is created and tested by launching
an European Association for Language Resources (ELRA), which takes care of distribution and
quality assurance of SLR, and which is linked to production centers of SLR.
The consortium is strongly positioned to create the infrastructure and exploit
the result because it represents the main European players in this field.
Participant Summary
- Siemens AG (Germany)
- GEC (GB)
- Jydsk Telefon (Denmark)
- Philips (Germany)
- CSELT (Italy)
- Telefones de Lisboa e Porto (Portugal)
- INESC (Inst. de Engenharia de Systemas e Computadores) (Portugal)
- Vocalis (GB)
- Dept of Phonetics UCL (GB)
- LIMSI-CNRS (France)
- Institute of Phonetic Science - Un. of Amsterdam (Netherlands)
- SPEX (Netherlands)
- Dep. de Filologia Espanyola - Un. Autonoma de Barcelona (Spain)
- Inst. fuer Phonetik und sprachliche Kommunikation/Univ. Muenchen (Germany)
- Personal Communication Lab.- Un. of Aalborg (Denmark)
- Institut de la Communication Parlée- INPG (France)
- Defence Research Agency (UK)
- IDIAP (Inst. Dalle Molle d'Intelligence Artificielle
Perceptive) (Switzerland)
Prime contractor
Harald Hoege (Siemens)
Siemens AG Abt. ZFE ST SN 53
Otto Hahn Ring 6
D81730 Muenchen
Germany
+49 89 6363374
+49 89 636 48000
e-mail: hoege@habicht.zfe.siemens.de