Adaptive phoneme alignment based on rough set theory

Abstract

The current work describes a phoneme matching algorithm based on rough set concepts. The objective of this type of algorithms is focused on the localization of the phonemic content of a specific spoken occurrence. According to the proposed algorithm, a number of rough sets containing the multiple expected phonemic instances in a sequence are created, each defined by a set of short term frames of the voice signal. The properties of the corresponding information system are derived from a features set calculated from the speech signal upon initiation. Given the above, an iterative procedure is applied by updating the phoneme instances versus the optimization of the accuracy metric. The main advantage of this algorithm is the absence of a training phase allowing for wider speaker adaptability and independency. The current paper focuses on the feasibility of the task as this work is still in early research stage. © 2010 Springer-Verlag Berlin Heidelberg.