A continuous speech recognition system integrating additional acoustic knowledge sources in a data-driven beam search algorithm

Abstract

The paper presents a continuous speech recognition system which integrates an additional acoustic knowledge source into the data-driven beam search algorithm. Details of the object oriented implementation of the beam search algorithm will be given. Integration of additional knowledge sources is treated within the flexible framework of Dempster-Shafer theory. As a first example, a rule-based plosive detector is added to the baseline system.