DocumentCode
3547779
Title
A speech recognizer with selectable model parameters
Author
Han, Wei ; Chan, Cheong-Fat ; Choy, Chiu-Sing ; Pun, Kong-Pang
Author_Institution
Dept. of Electron. Eng., Chinese Univ. of Hong Kong, Shatin, China
fYear
2005
fDate
23-26 May 2005
Firstpage
5842
Abstract
This paper presents the design and simulation results of a hidden Markov model (HMM) based isolated word recognizer IC. The new design can handle any combination of states and mixtures (up to 16 states and 8 mixtures). The speech IC has been verified with 353 test speech data. The recognition accuracy is 93.8% (48-bit) with no truncation and 88.9% with truncation (16-bit).
Keywords
hidden Markov models; speech recognition; 16 bit; 48 bit; CMOS; add-log function; hidden Markov model; look-up table; multimixture HMM; recognition accuracy; selectable model parameter speech recognizer; states/mixtures combination; truncation; word model; word recognizer IC; Character recognition; Design engineering; Hidden Markov models; Integrated circuit modeling; Integrated circuit testing; Pattern recognition; Probability density function; Probability distribution; Speech recognition; Viterbi algorithm;
fLanguage
English
Publisher
ieee
Conference_Titel
Circuits and Systems, 2005. ISCAS 2005. IEEE International Symposium on
Print_ISBN
0-7803-8834-8
Type
conf
DOI
10.1109/ISCAS.2005.1465967
Filename
1465967
Link To Document