Speech Recognition
Although the main focus of activity of the Department of Translation Technology (referred to by its French acronym TIM) has always been machine translation, the group has simultaneously maintained a strong interest in speech technology. Logically, things began with speech-to-speech translation; the group’s first project of this type was Spoken Language Translator (SLT; 1993-1999), a collaboration led by SRI International and Telia Research, Sweden, which produced one of the world’s first large-scale speech translation systems. The Department of Translation Technology (then ISSCO) was responsible for French speech and language processing.
The MedSLT and the CALL-SLT projects
Later, many of the ideas developed under SLT were reused to build the multilingual medical speech translation system MedSLT. The SNF-funded project ran from 2003 to 2008, and eventually produced a prototype system which could translate medical examination questions between French, English, Japanese, Spanish, Arabic and Catalan, using vocabularies of between 500 and 10000 words.
The Open Source toolkit which forms the base for MedSLT, Regulus, provides a set of general tools for constructing medium-vocabulary grammar-based spoken dialogue systems, and has since been used to build several other applications. The largest of these, CALL-SLT, adapts resources developed under MedSLT to create a platform for multilingual computer-assisted language learning, in which speech recognition is used to support interactive conversation practice between the computer and the student. Also funded by SNF, the first phase of this project started in 2009; a second phase began in 2014, and will continue until 2017. The main focus of activity concentrates on French, English and German.
- CALL-SLT demonstrators are available online.
The Regulus toolkit has also been exploited by groups outside TIM. In particular, it was used by NASA Ames Research Center to build Clarissa, an experimental astronaut assistant which in July 2005 became the first speech-enabled system to fly in space. Regulus is described in a 2006 book published by CSLI Press.
Main publications
- Putting Linguistics into Speech Recognition, Rayner M., Hockey B.A. and Bouillon P., in: CSLI, Stanford, 2006.
- The Spoken Language Translator, Rayner M., Carter D., Bouillon P., Digalakis V.and Wirén M. (eds), CUP:Cambridge, 2000.
- Pour une interlangue utile en traduction automatique de la parole dans des domaines limités , Bouillon P., Rayner M., Gerlach J., Estrella P., in: TAL (Traitement Automatique des Langues), Volume 52:1, 2011.
- Une grammaire partagée multi-tâche pour le traitement de la parole : application aux langues romanes , Bouillon P., Rayner M., Novellas Vall B., Starlander M., Santaholma M., Nakao Y. and Chatzichrisafis N., in: TAL (Traitement Automatique des Langues), Volume 47, 3/2006, Hermes and Lavoisier, 2007.
Consult a complete list of publications on speech recognition.