SPACE

Speech recognition techniques play an important part in the development of new user interfaces, and they would also be very useful in medical and educational applications involving an evaluation of the speaker's pronunciation, the speaker's reading ability, etc. Such applications involve tools for CALL (Computer Assisted Language Learning), reading tutors (for people with reading problems such as dyslexia) and pathological voice assessment tools (there is a need for objective scoring methods that offer alternatives to the irreproducible subjective scores one is utilizing now). What the mentioned applications have in common is that they are used by people with a non-standard speech output, and that the present automatic speech recognition algorithms experience problems with the correct processing of this output. Therefore the SPACE project will search for alternative solutions that work better for this kind of speech output. Among the ideas that will be explored by ELIS are: the investigation of articulatory features for characterizing a-typical speech (dysarthric speech, speech of the deaf, non-native speech), the development of pronunciation error models to detect pronunciation errors (misread words, interrupted words, etc.) and to give feedback on the type of error that was made. Within ELIS there'll be studied mainly objective quality measures for pathological speech by means of so-called articulatory feature detectors. Wrongly pronounced words and falters will also be examined on the lexical level using transformation rules. Project type: GBOU project by IWT - Flanders

Partners: 

ESAT-PSI (KULeuven)

ETRO-DSSP (VUB)

IWT Flanders