Publication Details
Czech Speech Recognizer for Multiple Environments
Karafiát Martin, Ing., Ph.D. (DCGM)
Burget Lukáš, doc. Ing., Ph.D. (DCGM)
Černocký Jan, prof. Dr. Ing. (DCGM)
speech, recognition, automatic, artiffical inteligence, training, czech, database, acoustic, modelling, modeling, language
This paper presents our work on building a largevocabulary continuous speech recognition (LVCSR) system for Czech, capable of op eration in multiple environments. SpeeCon and Temic speech databases were used to define a data-set for training acoustic models, attention was paid to unification of these two resources. The test set was also defined using these corp ora with careful choice of segments not overlapping with the training data. The system was completed by a language model trained on Czech National corpus. The recognition was performed using DUCoder an LVCSR stack decoder. Experimental results on the LVCSR task give a reference score of the system for future improvements.
@inproceedings{BUT22376,
author="Ondřej {Glembek} and Martin {Karafiát} and Lukáš {Burget} and Jan {Černocký}",
title="Czech Speech Recognizer for Multiple Environments",
booktitle="Radioeletronika 2006",
year="2006",
pages="1--4",
address="Bratislava",
url="http://www.fit.vutbr.cz/~glembek/papers/radioeletronika_2006.pdf"
}