An adaptive vocalic-phoneme learning model

摘要

The need of pronouncing specific vocalic phonemes correctly may arise when someone learns a foreign language or attends speech therapy sessions. A solution to this problem consists of using a computer-aided system that mirrors the phoneme learning process that presumably could show up on human learners, e.g. students and neurological patients. The proposed system is built up on a finite-state automata-based syntax-driven transducer and allows tracking any phonetic deviation by means of following specific state-and-transition pathways and updating percentage-based values on correct pronunciation for target phonemes. The employed adaptive pronunciation model makes use of a weight matrix to measure the degree of convergence/divergence in relation with the correct phoneme pronunciation. Thus, the inference mechanism can provide the human learner with appropriate phonetic-based pronunciation feedback as if it were a human teacher or therapist: it knows what has been learned and what must be learned. Consequently, both prediction of the human learner's pronunciation and a set of proposed words as didactic or therapeutic stimuli are delivered as well.