|without LM and DSG||34.1 %||4.6%|
|Table 2: Semantic interpretation of the segments|
The analysis of the segments forms the basis for the interpretation of the whole utterances, which we evaluated finally (see Table 3). In addition to the evaluation of the segment interpretation we introduce the value no, which denotes, that no interpretation was obtained, and the user was asked to repeat the instruction. More than 80 % of the utterances were interpreted correctly or partially correct, which demonstrates the strength of the approach.
|42.7 %||39.4 %||7.3 %||10.5%|
|Table 3: Interpretation of the utterances|
In this paper we presented an incremental speech recognition and interpretation system. The proposed architecture differs from traditional ones under two aspects: the knowledge is organised vertically and the modules work together in an integrated manner. The use ofa domain specific grammar improves the recognition results as well as the interpretation performance of the system.
( Eckert et al. 93) W. Eckert, T. Kuhn, H. Niemann, S. Rjeck, A. Scheuer, E. G. SchukatTalamazzini : A Spoken Dialogue System for German Intercity Train Timetable Inquiries. Proc. European Conf. on Speech Communication and Technology, Berlin, 1993, pp. 1871-1874.
( Lavie et al. 97) A. Lavie, A. Waibel, L. Levin, M. Finke, D. Gates, M. Gavalda, T. Zeppenfeld , P. Zhan,: "Janus III": Speech-to-Speech Translation in Multiple Languages. Proc. Int. Conf. on Acoustics, Speech and Signal Processing, Mütachen, 1997, pp. 99-104.
( McKevitt 94) P. McKevitt (ed.): AAAI-94 Workshop Program: Integration of Natural Language and Speech Processing, Seattle, Washington, 1994.
( Sagerer & Niemann 97) G. Sagerer, H. Niemann: Semalttic Neiworks for Understanding Scenes. Plenum Press, New York, 1997.
( Wachsmuth et al. 98) S. Wachsrnuth, G. A. Fink, G. Sagerer: Integration of Parsing and Incremental Speech Recognition. Proc. European Signal Processing Conf., Rhodes, 1998, pp. 371-374.