Presentation 1994/5/19
Recurrent neural networks for phoneme recognition
Takuya Koizumi, Shuji Taniguchi, Hidenori Ishida, Mikio Mori,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In this paper we propose six different recurrent neural networks- multi-layer networks with feedback architecture which are suitable for speech recognition.It is possible to attain higher recognition accuracies with the recurrent networks than those obtained with ordinary non-recurrent networks,because the recurrent networks can effectively learn dynamic variations of speech sound spectrum by an integrating or memorizing action of feedback loops with time delay within the networks.The superiority of the recurrent networks to non-recurrent networks has been confirmed by phoneme recognition experiments using a large number of speech samples.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) phoneme recognition / recurrent neural network / feedback architecture / multi-layered perceptron
Paper # SP94-1
Date of Issue

Conference Information
Committee SP
Conference Date 1994/5/19(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) Recurrent neural networks for phoneme recognition
Sub Title (in English)
Keyword(1) phoneme recognition
Keyword(2) recurrent neural network
Keyword(3) feedback architecture
Keyword(4) multi-layered perceptron
1st Author's Name Takuya Koizumi
1st Author's Affiliation Faculty of Engineering,Fukui University()
2nd Author's Name Shuji Taniguchi
2nd Author's Affiliation Faculty of Engineering,Fukui University
3rd Author's Name Hidenori Ishida
3rd Author's Affiliation Mitsubishi Electric
4th Author's Name Mikio Mori
4th Author's Affiliation Faculty of Engineering,Fukui University
Date 1994/5/19
Paper # SP94-1
Volume (vol) vol.94
Number (no) 42
Page pp.pp.-
#Pages 8
Date of Issue