Phoneme recognition with elliptic discrimination neural units

Noboru Kanedera

Ishikawa College of Technology, Ishikawa, Japan 929-03

Noboru Kanedera received a B.E. in Electronic Engineering from the University of Electro-communications, Tokyo, Japan, in 1985 and an M.E. in ElectronicEngineering from the University of Tokyo, Tokyo, in 1987. He is currently an Instructor at Ishikawa National College of Technology. His research interests include continuous speech recognitionand neural networks.

Search for more papers by this author

Tetsuo Funada,

Tetsuo Funada

Faculty of Technology, Kanazawa University, Kanazawa, Japan 920

Tetsuo Funada received a B.E. in Electronic Engineering from Kanazawa University, Kanazawa, Japan, in 1966, and an M.E. and a Dr. of Eng. degree in Electrical Engineering from Nagoya University, Nagoya, Japan, in 1968 and 1974, respectively. He is currently an Associate Professor on the Faculty of Technology at Kanazawa University. His research interests include speech pitch extraction, continuous speech recognition, and speech coding.

Search for more papers by this author

Noboru Kanedera,

Noboru Kanedera

Ishikawa College of Technology, Ishikawa, Japan 929-03

Search for more papers by this author

Tetsuo Funada,

Tetsuo Funada

Faculty of Technology, Kanazawa University, Kanazawa, Japan 920

Search for more papers by this author

First published: 1991

https://doi.org/10.1002/scj.4690221407

About

PDF

Tools

Share a link

Email
Wechat
Bluesky

Abstract

Many researchers achieved high phoneme recognition rates by multilayered neural networks with linear discrimination neural (LDN) units. However, it is difficult to analyze which components of the input are important to each unit in those LDN networks.

This paper proposed a multilayer neural network with elliptic discrimination neural (EDN) units so that the functions of each unit in the network may be interpreted more definitely. The center of the elliptic discrimination boundary of a neural unit corresponds to a typical point in an input space. The radii of the ellipse express the extent of the corresponding components in the input space, hence it becomes clear which components of the input space are important to each unit in the EDN network.

To compare the performance of EDN and LDN networks, recognition experiments of phonemes /b, d, g/ in 5240 tokens of a Japanese speech database were carried out. In the experiments, recognition rates were obtained by EDN networks as high as the rate by an LDN network. Also, it was confirmed which components of the input space are important to each unit in the EDN network.

References

1 A. Waibel, T. Hanazawa, G. Hinton, K. Shikano, and K. Lang. Phoneme recognition using time-delay neural networks. IEEE Trans. Acoust., Speech & Signal process., ASSP-37, 3, pp. 328–339 (1989).
10.1109/29.21701
Web of Science® Google Scholar
2 H. Kawahara and T. Iriro. A procedure for designing three-layer neural networks for pattern recognition applications. I.E.C.E. Tech. Rep., SP88-86, pp. 47–53 (1988).
Google Scholar
3 O. Saito, K. Uchimura, and A. Iwata. Analysis of internal representation in neural network consisting of higher-order neurons. Papers of Technical Group on Medical and Biological Engineering, I.E.C.E., MBE88-178, pp. 133–138 (1988).
Google Scholar
4 N. Kanedera and T. Funada. Phoneme recognition of continuous speech using higher-order neural networks. In: Proc. Fall Meet. Acoust. Soc. Japan (Oct. 1989).
Google Scholar
5 N. Kanedera and T. Funada. /b, d, g/ recognition with elliptic discrimination neuron. In: Proc. Spring Meet. Acoust. Soc. Japan (March 1990).
Google Scholar
6 K. Takeda, Y. Sagisaka, S. Katagiri, and H. Kuwabara. Construction of an acoustically phonetically transcribed Japanese speech database. I.E.C.E. Tech. Rep., SP87-19, pp. 25–32 (1987).
Google Scholar
7 D. E. Rumelhart, G. E. Hinton, and R. J. Williams. Learning representations by back-propagating errors. Nature, 323, pp. 533–536 (Oct. 1986).
10.1038/323533a0
Web of Science® Google Scholar
8 N. Kanedera and T. Funada. A neural network for phonetic segmentation of continuous speech. Trans. I.E.I.C.E., Japan, J73-D-II, No. 1, pp. 72–79 (1990).
Google Scholar
9 S. Doshita, T. Kawahara, Y. Mizutani, H. Kojima, M. Ishikawa, and S. Kitazawa. Speaker-independent discrimination of Japanese consonants in isolated syllables using pairwise discrimination method. The Journal of the Acoust. Soc. Japan, 45, No. 11, pp. 827–836 (June 1989).
Google Scholar
10 K. Tujino, T. Sakurai, Y. Nomura, S. Chigusa, R. Mizoguchi, and O. Kakusho. A continuous speech recognition system with a powerful environment for the knowledge-base construction. Trans. I.E.I.C.E., Japan, J71-D, No. 3, pp. 531–542 (1988).
Google Scholar

Volume22, Issue14

1991

Pages 90-100

Phoneme recognition with elliptic discrimination neural units

Abstract

References

References

Information

About Wiley Online Library

Help & Support

Opportunities

Connect with Wiley

Phoneme recognition with elliptic discrimination neural units

Abstract

References

References

Related

Information