A consideration of learning in speech recognition from the viewpoint of AI class-description learning

Abstract

The learning mechanism used in a user-adaptive speech recognizer based on the subspace method is treated. Comparing the subspace learning system with the AI (artificial intelligence) learning system ARCH, the following points are made: (1) subspace learning using covariance matrix modification and KL-expansion is a kind of class-description learning, as found in ARCH. The subspace method focuses on feature extraction for powerful pattern class representation, but does not involve only pattern classification; (2) the concept of near-miss in ARCH can be simulated with the subspace method; (3) M. Minsky's recent (1985) concept 'uniframe', which represents a meaning of a class, is obtained as a subspace with KL-expansion.<>