Conference paper

 
Grézl, F.: The Role of Neural Network Size in TRAP/HATS Feature Extraction, In: Proceedings Text, Speech and Dialogue 2011, Plzeň, CZ, Springer, 2011, p. 315-322, ISBN 978-3-642-23537-5, ISSN 0302-9743
Publication language:english
Original title:The Role of Neural Network Size in TRAP/HATS Feature Extraction
Title (cs):Role velikosti neuronové sítě v extrakci příznaků pomocí TRAP/HATS
Pages:315-322
Proceedings:Proceedings Text, Speech and Dialogue 2011
Conference:14th International Conference on Text, Speech and Dialogue
Series:LNAI 6836
Place:Plzeň, CZ
Year:2011
ISBN:978-3-642-23537-5
Journal:Lecture Notes in Computer Science, Vol. 2011, No. 9, DE
ISSN:0302-9743
Publisher:Springer Verlag
URL:http://www.fit.vutbr.cz/research/groups/speech/publi/2011/grezl_tsd2011.pdf [PDF]
Keywords
Neural networks, feature extraction, probabilistic features
Annotation
This article examines the performance of TRAP/HATS based probabilistic features in ASR. The sizes of neural networks in both stages of processing are changed and the influence is evaluated.
Abstract
We study the role of sizes of neural networks (NNs) in TRAP (Tempo- RAl Patterns) and HATS (Hidden Activation TRAPS architecture) probabilistic features extraction. The question of sufficient size of band NNs is linked with the question whether the Merger is able to compensate for lower accuracy of band NNs. For both architectures, the performance increases with increasing size of Merger NN. For TRAP architecture, it was observed, that increasing band NN size over some value has not further positive effect on final performance. The situation is different when HATS architecture is employed - increasing size of band NNs has mostly negative effect on final performance. This is caused by merger not being able to efficiently exploit the information hidden in its input with increased size. The solution is proposed in form of bottle-neck NN which allows for arbitrary size output.
BibTeX:
@INPROCEEDINGS{
   author = {František Grézl},
   title = {The Role of Neural Network Size in TRAP/HATS Feature
	Extraction},
   pages = {315--322},
   booktitle = {Proceedings Text, Speech and Dialogue 2011},
   series = {LNAI 6836},
   journal = {Lecture Notes in Computer Science},
   volume = {2011},
   number = {9},
   year = {2011},
   location = {Plzeň, CZ},
   publisher = {Springer Verlag},
   ISBN = {978-3-642-23537-5},
   ISSN = {0302-9743},
   language = {english},
   url = {http://www.fit.vutbr.cz/research/view_pub.php?id=9751}
}