Memory efficient and fast speech recognition system for lowresource mobile devices

Author

Chung, Hoon ; Chung, Ikjoo

Author_Institution

Spoken Language Process. Team, ETRI

Volume

52

Issue

3

fYear

2006

Firstpage

792

Lastpage

796

Abstract

In this paper, we consider practical issues such as memory efficiency and fast decoding to make continuous density hidden Markov model (CDHMM)-based large vocabulary speech recognition system work on resource limited mobile devices. Particularly, we focus on memory efficient acoustic modeling and fast state likelihood computation. The proposed techniques are implemented in a speaker-independent Korean speech recognition system running on a personal digital assistant (PDA) with a 32-bit fixed-point processor operating at 400 MHz. The system uses 0.5 MB memory for representing 28448 Gaussians and it runs at 2.54xRT without serious degradation of accuracy on 10 k phonetically optimized words recognition task domain

Keywords

decoding; hidden Markov models; mobile handsets; natural languages; notebook computers; speaker recognition; speech coding; 0.5 MB; 400 MHz; PDA; acoustic modeling; continuous density hidden Markov model; decoding; fast state likelihood computation; fixed-point processor; low-resource mobile devices; memory efficient; personal digital assistant; speaker-independent Korean speech recognition system; vocabulary speech recognition system; Acoustic devices; Computer interfaces; Decoding; Hidden Markov models; Keyboards; Mobile computing; Personal digital assistants; Probability; Speech recognition; Vocabulary;

fLanguage

English

Journal_Title

Consumer Electronics, IEEE Transactions on

Publisher

ieee

ISSN

0098-3063

Type

jour

DOI

10.1109/TCE.2006.1706471

Filename

1706471