278 - Spoken Language Interaction in Telecommunication. Chair: B. Lindberg, Denmark

The objectives of the Action were summarized in 5 points:

– To improve the knowledge of the issues and problems involved in general in spoken language interaction in telecommunication.
– To achieve knowledge of issues related to robustness and multi-linguality within spoken language processing.
– To achieve knowledge of spoken language interaction in the context of multimodal communication.
– To achieve knowledge of human-computer dialogue theories, models and systems and associated tools for the establishment of such systems.
– To achieve knowledge of and evaluate telecommunication applications, which apply spoken language as one out of more input or output modalities.

A multilingual Broadcast News database covering 9 languages was established by the Action. Software to enable uniform evaluation of front-end algorithms from various groups was developed. Also, segmentation and classification algorithms from 5 groups were evaluated. Short term scientific missions have taken place in connection with the annotation, segmentation and evaluation of the recorded database, fostering collaboration among research groups across several countries. A consortium was established in order to continue collaboration on Broadcast News after the end of the COST278. Furthermore, collaborative work between seven research groups was performed within cross-lingual and multi-lingual speech recognition. Specific results on using acoustic data for one language in the acoustic modelling of another language were reported. Here, a data-driven and expert-driven method were compared in obtaining the best acoustic model for the Slovenian language, based on speech data for the German, Spanish, Hungarian, and Slovak languages.