Parametric subspace modeling of speech transitions

This report describes an attempt at capturing segmental transition information for speech recognition tasks. The slowly varying dynamics of spectral trajectories carries much discriminant information that is very crudely modelled by traditional approaches such as HMMs. In approaches such as recurrent neural networks there is the hope, but not the convincing demonstration, that such transitional information could be captured. The method presented here starts from the very diierent position of explicitly capturing the trajectory of short time spectral parameter vectors on a subspace in which the temporal sequence information is preserved. We approach this by introducing a temporal constraint into the well known technique of Principal Component Analysis. On this subspace, we attempt a parametric modelling of the trajectory, and compute a distance metric to perform classiication of diphones. We use the principal curves method of Hastie and Stuetzle and the Generative Topographic map (GTM) technique of Bishop, Svenson and Williams to describe the temporal evolution in terms of latent variables. On the diicult problem of /bee/, /dee/, /gee/ we are able to retain discriminatory information with a small number of parameters. Experimental illustrations present results on ISOLET and TIMIT database. ISOLET und TIMIT experimentell illustriert, die in den Bericht integriert sind.