Presentation 1997/9/11
An Error Correction Method with HMMs based on Maximum Likelihood Estimation
Tadashi Yonezaki, Koji Yoshida, Toshio Yagi,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) We propose a new error correction method using Hidden Markov Models (HMMs). The code sequence is decoded by a maximum likelihood estimation (ML). The measure for the ML consists of two probabilities extracted from a source model specified with HMMs and a channel model. Therefore this approach doesn't require any forward error control codes. The proposed approach is applied to correct a code sequence of quantized LSP parameters. Over a noisy channel which is simulated by adding 3% bit errors randomly, it realizes about 39% reduction of cepstrum distortion (16 states HMMs).
Keyword(in Japanese) (See Japanese page)
Keyword(in English) error correction / HMMs / source model / channel model / maximum likelihood estimation
Paper # SP97-43
Date of Issue

Conference Information
Committee SP
Conference Date 1997/9/11(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Speech (SP)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) An Error Correction Method with HMMs based on Maximum Likelihood Estimation
Sub Title (in English)
Keyword(1) error correction
Keyword(2) HMMs
Keyword(3) source model
Keyword(4) channel model
Keyword(5) maximum likelihood estimation
1st Author's Name Tadashi Yonezaki
1st Author's Affiliation Telecom Research Lab., Matsushita Communication Ind. Co., Ltd.()
2nd Author's Name Koji Yoshida
2nd Author's Affiliation Telecom Research Lab., Matsushita Communication Ind. Co., Ltd.
3rd Author's Name Toshio Yagi
3rd Author's Affiliation Telecom Research Lab., Matsushita Communication Ind. Co., Ltd.
Date 1997/9/11
Paper # SP97-43
Volume (vol) vol.97
Number (no) 248
Page pp.pp.-
#Pages 7
Date of Issue