A Semantic and Detection-based Approach to Speech and Language Processing

  • Li Deng ,
  • KS Wang ,
  • R. Guido

in Phillip Sheu; Heather Yu; C V Ramamoorthy; Arvind K Joshi; Lotfi A Zadeh (eds) Semantic Computing

Published by Wiley | 2010

This chapter presents a new formulation that tightly integrates the detection – based algorithm into the maximum a posteriori (MAP) decision. The key to this formulation is to implement the sequential detection algorithm and to recurrently apply the sequential probability ratio test in a time – synchronous, single – pass decoding framework. The chapter shows that realizing the detection – based recognition in single – pass architecture is feasible. It provides an overview of the mathematical foundation of this approach, serving as an introduction to the general detection – based approach for computer processing of speech and language. This overview starts with the conventional fixed – sample – size detection, which then naturally extends to sequential detection theory. Finally, it presents a comprehensive case study on how the sequential detection technique is successfully applied to a speech understanding task that is related to personal information management.