DocumentCode
339169
Title
Correlation generated targets for neural network based speech recognition
Author
Li, Fei ; Yan, Yonghong
Author_Institution
Nanjing Inst. of Post & Telecommun., China
fYear
1998
fDate
1998
Firstpage
718
Abstract
This paper presents a novel technique for data sharing through neural network training target manipulation. It uses correlations between output values of a network to create a set of non-zero targets for each training pattern. Experimental results show that more than 20% error reduction is achieved on a speaker-independent continuous digit string recognition task
Keywords
correlation methods; learning (artificial intelligence); neural nets; speech recognition; correlation generated targets; data sharing; error reduction; neural network based speech recognition; neural network training target manipulation; non-zero targets; output values; recognition task; speaker-independent continuous digit string; training pattern; Acoustic measurements; Context modeling; Hidden Markov models; Least squares approximation; Natural languages; Neural networks; Robustness; Speech recognition; Telephony; Training data;
fLanguage
English
Publisher
ieee
Conference_Titel
Signal Processing Proceedings, 1998. ICSP '98. 1998 Fourth International Conference on
Conference_Location
Beijing
Print_ISBN
0-7803-4325-5
Type
conf
DOI
10.1109/ICOSP.1998.770312
Filename
770312
Link To Document