HTK is a portable software toolkit for building speech recognition systems using continuous density hidden Markov models developed by the Cambridge University Speech Group. One particularly successful type of system uses mixture density tied-state triphones. We have used this technique for the 5 k/20 k word ARPA Wall Street Journal (WSJ) task. We have extended our approach from using word-internal gender independent modelling to use decision tree based state clustering, cross-word triphones and gender dependent models. Our current systems can be run with either bigram or trigram language models using a single pass dynamic network decoder. Systems based on these techniques were included in the November 1993 ARPA WSJ evaluation, and gave the lowest error rate reported on the 5 k word bigram, 5 k word trigram and 20 k word bigram “hub” tests and the second lowest error rate on the 20 k word trigram “hub” test
%0 Conference Paper
%1 Woodland1994
%A Woodland, Phil C.
%A Odell, Julian J.
%A Valtchev, Valtcho
%A Young, Steve J.
%B Proceedings of the 1994 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
%C Adelaide, Australia
%D 1994
%K Group;HTK;bigram Journal Markov Speech Street University Wall analysis;Hidden based clustering;error data;Vocabulary decision decoder;speech density dependent dynamic gender hidden independent language languages;software modelling;Buildings;Decision models;Software models;continuous models;mixture models;natural models;word-internal network pass rate;gender recognition recognition;System recognition;cross-word recognition;vocabulary;ARPA software speech state systems;trigram task;Cambridge testing;Training theory;decoding;hidden tied-state toolkit;single tools;Speech tools;speech tree trees;Decoding;Error triphones;decision triphones;portable
%P 125-128
%R 10.1109/ICASSP.1994.389562
%T Large vocabulary continuous speech recognition using HTK
%V 2
%X HTK is a portable software toolkit for building speech recognition systems using continuous density hidden Markov models developed by the Cambridge University Speech Group. One particularly successful type of system uses mixture density tied-state triphones. We have used this technique for the 5 k/20 k word ARPA Wall Street Journal (WSJ) task. We have extended our approach from using word-internal gender independent modelling to use decision tree based state clustering, cross-word triphones and gender dependent models. Our current systems can be run with either bigram or trigram language models using a single pass dynamic network decoder. Systems based on these techniques were included in the November 1993 ARPA WSJ evaluation, and gave the lowest error rate reported on the 5 k word bigram, 5 k word trigram and 20 k word bigram “hub” tests and the second lowest error rate on the 20 k word trigram “hub” test
@inproceedings{Woodland1994,
abstract = {HTK is a portable software toolkit for building speech recognition systems using continuous density hidden Markov models developed by the Cambridge University Speech Group. One particularly successful type of system uses mixture density tied-state triphones. We have used this technique for the 5 k/20 k word ARPA Wall Street Journal (WSJ) task. We have extended our approach from using word-internal gender independent modelling to use decision tree based state clustering, cross-word triphones and gender dependent models. Our current systems can be run with either bigram or trigram language models using a single pass dynamic network decoder. Systems based on these techniques were included in the November 1993 ARPA WSJ evaluation, and gave the lowest error rate reported on the 5 k word bigram, 5 k word trigram and 20 k word bigram “hub” tests and the second lowest error rate on the 20 k word trigram “hub” test},
added-at = {2021-02-01T10:51:23.000+0100},
address = {Adelaide, Australia},
author = {Woodland, Phil C. and Odell, Julian J. and Valtchev, Valtcho and Young, Steve J.},
biburl = {https://www.bibsonomy.org/bibtex/2ebd3370bf8762b58ecd4220296e68784/m-toman},
booktitle = {Proceedings of the 1994 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
doi = {10.1109/ICASSP.1994.389562},
file = {:pdfs/woodland_icassp_1994.pdf:PDF},
interhash = {61bf4dab4511cce144c07b2a6730c3fe},
intrahash = {ebd3370bf8762b58ecd4220296e68784},
issn = {1520-6149},
keywords = {Group;HTK;bigram Journal Markov Speech Street University Wall analysis;Hidden based clustering;error data;Vocabulary decision decoder;speech density dependent dynamic gender hidden independent language languages;software modelling;Buildings;Decision models;Software models;continuous models;mixture models;natural models;word-internal network pass rate;gender recognition recognition;System recognition;cross-word recognition;vocabulary;ARPA software speech state systems;trigram task;Cambridge testing;Training theory;decoding;hidden tied-state toolkit;single tools;Speech tools;speech tree trees;Decoding;Error triphones;decision triphones;portable},
month = apr,
owner = {schabus},
pages = {125-128},
timestamp = {2021-02-01T10:51:23.000+0100},
title = {Large vocabulary continuous speech recognition using HTK},
volume = 2,
year = 1994
}