Skip to content

Detail of publication


Šmídl, L. and Psutka Josef V. and Zahradil, J. : Keyword spotting with triphone based filler model . SPECOM 2005 proceedings, p. 487-490, Moscow State Linguistics University, Moscow , 2005.


This paper deals with implementing a triphone based filler model into our keyword spotting system. The presented keyword spotting (KWS) algorithm is designed for telephone applications, spontaneous speech and it can be used with an advantage in situation when no proper language model is available. The equal error rate of the triphone filler model (FM) system (including block decision) is 17.3% which is 43.8% relative improvement over the baseline system (monophone based filler model). The FOM (figure of merit) rises up from 87.80% (monophone FM) to 94.69% (triphone FM). Furthermore, the block decision is introduced. The decision is not carried out in each decoding time frame but only when the record stored in the last keyword state is changed. This approach allows us to use more information for decision – besides a score of the last keyword state also additional values, e.g. length of the keyword, could be considered to support the decision.

Detail of publication

Title: Keyword spotting with triphone based filler model
Author: Šmídl, L. ; Psutka Josef V. ; Zahradil, J.
Language: English
Date of publication: 17 Oct 2005
Year: 2005
Type of publication: Papers in proceedings of reviewed conferences
Title of journal or book: SPECOM 2005 proceedings
Page: 487 - 490
ISBN: 5-7452-0110-X
Publisher: Moscow State Linguistics University
Address: Moscow
Date: 17 Oct 2005 - 19 Oct 2005
/ 2011-06-09 13:08:53 /


keyword spotting, acoustic baseform, speech recognitio


 author = {\v{S}m\'{i}dl, L. and Psutka Josef V. and Zahradil, J.},
 title = {Keyword spotting with triphone based filler model},
 year = {2005},
 publisher = {Moscow State Linguistics University},
 journal = {SPECOM 2005 proceedings},
 address = {Moscow },
 pages = {487-490},
 ISBN = {5-7452-0110-X},
 url = {},