A System for Improving the Recognition of Fluently Spoken German Speech

A research p ro jec t f o r improving the recogni t i o n of f l u e n t l y spoken German speech is presented. The work is in progress at present . * It should be i nves t i ga ted , how fa r aspects of semantics and inferences could improve the automat i c speech recogn i t i on . The work is part of a speech recogn i t ion system that receives speech s i g na l s , converts them in to forms su i tab le fo r f u r t he r act ions and f i n a l l y puts out the spoken tex t in characters. The system i t s e l f operates at three stages. Within the f i r s t one the signal analys is is performed using a well-known method. This analys is segments the signal i n to ce r ta in subword un i ts and, f o r each segment, produces a set of weighted candi dates. At the second stage these candidates are used to generate weighted word hypotheses w i th the a id of an extensive l ex i con . The hypotheses have to be v e r i f i e d or f a l s i f i e d w i t h i n the fo l l ow ing processing steps at the t h i r d stage. Thereby the a lgor i thm uses a b e s t f i r s t s t rategy (hypotheses wi th highest weight f i r s t ) . Besides syntact ic/grammatical aspects, semant i c analys is and inferences mentioned above are the methods, tha t should lead to a ce r ta in text-comprehension.