Speech Translation for French in the NESPOLE! European Project

摘要

This paper presents CLIPS laboratory activities in the context of the NESPOLE! European project, exploring future applications of automatic speech to speech translation in e-commerce and e-service sectors. The scientific and technological research issues particularly addressed in order to improve current experimental speech-to-speech translation systems, are: robustness, scalability, and cross-domain portability. The general architecture of the whole speech to speech translation demonstrator is first presented and the Interchange Format (IF) strategy for translation adopted in the project is quickly described. The French database recorded during the project and the French Human Language Technology (HLT) modules (recognition, synthesis and translation) are then fully detailed. First results obtained and future perspectives of the project are also discussed in this article.