Research objectives and content
The development of mathematical models of speech motor control is important in understanding and representing sources of phonetic variability during speech production. Although databases of articulatory measurements are available, and detailed physiological models have been proposed, little progress has been made in evaluating competing model hypotheses on experimental data, due to the lack of appropriate mathematical estimation procedures for fitting model parameters automatically to physical measurements. The research described in this proposal aims to develop a statistical framework for training, testing, and analysing stochastic models of speech production using physical data. Algorithms for recovering model parameters and control variable trajectories directly from articulatory measurements will be derived, drawing on similar results in speech recognition developed as part of the applicant's doctoral thesis. The results will be applied to the equilibrium-point model of motor control developed by the host institute, and evaluated on corpora of articulatory data.
Training content (objective, benefit and expected impact)
The applicant will gain experience in constructing computational models of motor control, and will receive training in the associated experimental methodology. The project will be of considerable benefit in establishing a useful collaboration between British, French, and Canadian researchers in speech production and recognition.
Links with industry / industrial relevance (22)