Computational models for speech production

From MaRDI portal





The aim of this tutorial paper is to survey, to analyze, and to compare the major classes of statistical dynamic speech production models, with their theoretical motivation, mathematical formulation, and procedures for parameter learning. A particular type of statistical task-dynamic model of speech production is developed which integrates the strengths of the scientific and technological styles of the current production models. Both maximum likelihood and minimum classification error criteria are used for model parameter learning, justified in terms of statistical optimization principles which are integrated in neural networks and dynamic system theories. The optimization of the likelihood or empirical classification error rate for a small number of hyper-parameters in the proposed speech production model permits the robust modelling of true dynamic behaviours of human speech. For the entire collection see [Zbl 0927.00044].











This page was built for publication: Computational models for speech production

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4942744)