An automatic acquisition method of statistic finite-state automaton for sentences

Author

Suzuki, Motoyuki ; Makino, Shozo ; Aso, Hirotomo

Author_Institution

Comput. Center, Tohoku Univ., Sendai, Japan

Volume

2

fYear

1999

fDate

15-19 Mar 1999

Firstpage

737

Abstract

Statistic language models obtained from a large number of training samples play an important role in speech recognition. In order to obtain higher recognition performance, we should introduce long distance correlations between words. However, traditional statistic language models such as word n-grams and ergodic HMMs are insufficient for expressing long distance correlations between words. We propose an acquisition method for a language model based on HMnet taking into consideration long distance correlations and word location

Keywords

correlation methods; finite automata; hidden Markov models; natural languages; speech recognition; statistical analysis; HMnet; artificial language; automatic acquisition method; ergodic HMM; language model; long distance correlations; natural language; recognition performance; sentences; speech recognition; statistic finite-state automaton; statistic language models; training samples; word location; word n-grams; Automata; Automatic speech recognition; Dictionaries; Hidden Markov models; Natural languages; Probability distribution; Speech recognition; State estimation; Statistics;

fLanguage

English

Publisher

ieee

Conference_Titel

Acoustics, Speech, and Signal Processing, 1999. Proceedings., 1999 IEEE International Conference on

Conference_Location

Phoenix, AZ

ISSN

1520-6149

Print_ISBN

0-7803-5041-3

Type

conf

DOI

10.1109/ICASSP.1999.759772

Filename

759772