5th European Conference on Speech Communication and Technology

Rhodes, Greece
September 22-25, 1997

Improvement on Connected Digits Recognition Using Duration Constraints in the Asynchronous Decoding Scheme

Miroslav Novak

IBM Watson Research Center - Human Language Technologies Group, Yorktown Heights, NY, USA

This paper describes the use of an explicit word duration model in the environment of a HMM based time asynchronous stack search decoder. The benefit of the method is demonstrated on the task of connected digit recognition. Analysis of typical errors observed on this task suggests that appropriate word duration modeling can improve recognition accuracy. Duration model based on the Gamma Distribution, applied as a post- processing step during iterations of the search algorithm, reduces the error rate of the baseline system by 14%.

Full Paper

Bibliographic reference.  Novak, Miroslav (1997): "Improvement on connected digits recognition using duration constraints in the asynchronous decoding scheme", In EUROSPEECH-1997, 159-162.