Speech Modelling Using Subspace and EM Techniques