Phonetic Feature Recognition with Time Delay Neural Network and the Evaluation by Mutual Information
-
- Okawa Shigeki
- School of Science and Engineering,Waseda University
-
- Windheuser Christoph
- Telecom Paris,Department Signal,C.N.R.S-URA820
-
- Bimbot Frederic
- Telecom Paris,Department Signal,C.N.R.S-URA820
-
- Shirai Katsuhiko
- School of Science and Engineering,Waseda University
Bibliographic Information
- Other Title
-
- TDNNによる音韻の弁別特徴認識とその相互情報量による評価
Search this article
Description
In this paper,we describe a TDNN based hybrid word recognition system with a novel phoneme representation based on phonetic features.This new representation is more compact than the standard phoneme representation and leads to a fast training of networks. Furthermore,we introduce a mutual information as a new measure to evaluate the different characterized networks.With this measure,we perform the optimization of a modular TDNN.In different experiments on a spelled letter recognition,we show that the network using the set of phonetic features which optimized by the mutual information leads to better recognition result.
Journal
-
- IEICE technical report. Speech
-
IEICE technical report. Speech 93 (427), 25-32, 1994-01-21
The Institute of Electronics, Information and Communication Engineers
- Tweet
Keywords
Details 詳細情報について
-
- CRID
- 1570291227409839616
-
- NII Article ID
- 110003296781
-
- NII Book ID
- AN10013221
-
- Text Lang
- ja
-
- Data Source
-
- CiNii Articles