Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

Speed Improvement of the Tree-Based Time Asynchronous Search

Miroslav Novak, Michael Picheny

IBM T.J. Watson Research Center, Yorktown Heights, NY, USA

The IBM large vocabulary continuous speech recog- nition system is based on an asynchronous stack de- coding scheme. This is essentially a tree search, as described in [1]. The main advantages - ecient mem- ory utilization and a single-pass search strategy - make the system extremely suitable for real-time applica- tions. This article describes further improvements in eciency of the search method. These improvements are achieved in part by more ecient word to con- text dependent acoustic model expansion, producing equivalent search results and thus not affecting the recognition accuracy. Additional improvements are achieved by introducing an approximation in the com- putation of the likelihood of the hypothesized path. The basic idea is to allow sharing of some branches in the search tree and results in effectively a tree to network transformation.


  1. Rabiner, L.R., Juang, B.H., \An introduction to hidden Markov Models", IEEE ASSP Mag., Vol. 3, pp. 4-16, Jan. 1986

Full Paper

Bibliographic reference.  Novak, Miroslav / Picheny, Michael (2000): "Speed improvement of the tree-based time asynchronous search", In ICSLP-2000, vol.4, 334-337.