Introduction

This is a preliminary version of the Vystadial 2013 acoustic data & scripts dataset. This version is not an official release, it serves only as a support for the paper submission for `LREC 2014`_.

The abstract of the paper describing the data and related scripts is in the file abstract.pdf.

The data and scripts are found in directories as follows:

data_voip_cs/{train,dev,test}

the Czech data

data_voip_en/{train,dev,test}

the English data

htk

scripts for HTK

kaldi

scripts for Kaldi

Data statistics

Latest statistics for the data are as follows:

dataset	audio	#sents	#words
English
train	19:52	22,933	103,136
dev	1:44	2,000	9,174
test	1:43	2,000	8,970
Czech
train	10:57	15,319	93,177
dev	1:24	2,000	11,854
test	1:24	2,000	11,841

Licence for the data

See LICENSE-CC-BY-SA-3.0.TXT.

Licence for the training scripts

See LICENSE-APACHE-2.0.TXT.

Authors

• Matěj Korvas
• Ondřej Plátek
• Ondřej Dušek
• Lukáš Žilka
• Filip Jurčíček