Publication Details
Integrating recent MLP feature extraction techniques into TRAP architecture
TRAP processing, Bottle-Neck technique, subphoneme classes, LVCSR features
The article shows the performance improvement and limitations of TRAP and HATS neural net systems for feature extraction in LVCSR when the bottel neck approach and phoneme states targets are introduced into them.
This paper is focused on the incorporation of recent techniques for multi-layer perceptron (MLP) based feature extraction in Temporal Pattern (TRAP) and Hidden Activation TRAP (HATS) feature extraction scheme. The TRAP scheme has been origin of various MLP-based features some of which are now indivisible part of state-of-the-art LVCSR systems. The modifications which brought most improvement - sub-phoneme targets and Bottle-Neck technique - are introduced into original TRAP scheme. Introduction of sub-phoneme targets uncovered the hidden danger of having too many classes in TRAP/HATS scheme. On the other hand, Bottle-Neck technique improved the TRAP/HATS scheme so its competitive with other approaches.
@INPROCEEDINGS{FITPUB9755, author = "Franti\v{s}ek Gr\'{e}zl and Martin Karafi\'{a}t", title = "Integrating recent MLP feature extraction techniques into TRAP architecture", pages = "1229--1232", booktitle = "Proceedings of Interspeech 2011", journal = "Proceedings of Interspeech - on-line", volume = 2011, number = 8, year = 2011, location = "Florence, IT", publisher = "International Speech Communication Association", ISBN = "978-1-61839-270-1", ISSN = "1990-9772", language = "english", url = "https://www.fit.vut.cz/research/publication/9755" }