An investigation into articulatory vocoding for vowels, as a means of achieving high quality coding at low bit rates, is carried out in this thesis. Methods of estimating the vocal tract transfer function from the speech wave are compared, and an algorithm for closed glottis interval (CGI) analysis is developed. CGI analysis is chosen over autocorrelation based inverse filtering methods.
Various distortion measures for use in Vector Quantization are evaluated, and a new covariance distortion measure is proposed. It is shown that this measure yields close matches from an acoustic codebook.
An articulatory coding system is designed, including a linked codebook of articulatory shapes, based on synthetic speech. A method of generating a similar codebook from real speech is proposed, and an investigation into estimating articulatory parameters from the speech wave is carried out to this end.