A structured speech model with continuous hidden dynamics and prediction-residual training for tracking vocal tract resonances | IEEE Conference Publication | IEEE Xplore