Third International Conference on Spoken Language Processing (ICSLP 94)
A model for lexical analysis based on reliable information stemming from the acoustic-phonetic decoder is presented in this paper. Reliable information is conveyed by some of the acoustic cues used for the acoustic-phonetic decoding which are both particularly discriminant and well pronounced at the time of realization. Lexical access and lexical analysis we put forward also aim at maintaining the consistency of the overall process keeping track of the context in which identification hypotheses have been made. We describe the lexical module of the system DAPHNE designed for French sentences made of stops and vowels. We also present acoustic cues used on stops followed by back vowels and give preliminary results obtained on these sounds.
Bibliographic reference. Coste-Marquis, Sylvie (1994): "Interaction between most reliable acoustic cues and lexical analysis", In ICSLP-1994, 2187-2190.