| Popis: |
This study deals with a specific speech phenomenon, anticipation, a suprasegmental feature of speech utterance that allows an interpreter to predict, what the speaker has in his/her mind and how his/her speech will continue in real time. Based on extensive perceptual testing, a speech corpus containing anticipation foci was created. These data patterns of the nucleus were statistically analysed and obtained parameters show how suprasegmental characteristics of anticipation should be effectively analysed. Two systems were developed for anticipation nucleus detection, Hidden Markov Models and another based on fuzzy logic. These approaches were trained, and then their detection score was evaluated. Proposed detection algorithms and statistical characteristics of anticipatory foci can be applied in interpreting support equipment or in the process of interpreting education and also used for the improvement of a speech recognition system, a dialogue management system in a human-machine interaction. |
