Using prosodic rules in speech recognition expert system

A description is given of the use of prosodical rules for segmenting speech signals in sentences and words (lexical and grammatical words) and putting milestones on their boundaries. Prosodical parameters (pitch, energy, duration) are estimated over vocalic nuclei and normalized by using uniform coding. The supervisor in an automatic speech recognition expert system is guided by these milestones. The rules are inferred by statistical learning of prosodical parameters and are then assessed linguistically by an expert.<>