Uni
Faculties
Fk. V
Physics
Medical Physics
Download
Download of the Oldenbug Logatom Corpus (OLLO)
The Oldenburg Logatome Corpus is a multi-lingual speech
database, which was built within the research
project DIVINES (Diagnostic and Intrinsic Variabilities in Natural
Speech) supported by the European Commission (FP6-002034). One of the
aims of this project is the comparison of human and automatic speech recognition.
The OLLO corpus is specifically designed for this task and consists of
CVC (consonant vowel consonant) and VCV (vowel consonant vowel) logatomes
(total 150 different logatomes) with the same initial and final character.
The corpus is composed of read speech from 50 speakers (40 German and
10 French speakers) including speaker dependent and speaker independent
speech variabilities. These include regional dialect, gender, age as well
as speaking rate (fast, normal, slow), speaking effort (soft, normal,
raised), and speaking style (statement, question). The database was presented
on the Interspeech conference 2005;
the paper can be downloaded here.
On this website, you can download a description of the
OLLO corpus and the database itself. Each package contains recordings
from all speakers from a dialect region.
Important notice: The OLLO speech database is subject of further
development and enhancement. We recommend that you sign up for our OLLO
newsletter, so we can inform you about updated versions and improvements
concerning the corpus. In order to sign up, just click the following link
and enter your mail address as message text: Sign
up for OLLO newsletter (please replace _AT_ in the mail address with
@). A version history can
be found here.
Copyright notice: Permission to use this database
for purely research or educational
purposes is granted. No commercial exploitation of this database is permitted
unless permission has been obtained separately from UNIVERSITAET OLDENBURG
(contact adress: medi-ollo_AT_listserv.uni-oldenburg.de).
Copyright 2005, Medizinische Physik, Universitaet Oldenburg, Germany.
All rights reserved.
 |
OLLO2.0_README.ZIP |
Corpus description
(6 kB) |
| |
|
OLLO2.0_NO.ZIP |
Audio data
package 1/5: German speakers (1 - 10), no dialect (846 MB) |
| |
|
OLLO2.0_EF.ZIP |
Audio data
package 2/5: German speakers 11 - 20, East Frisian dialect (898 MB) |
| |
|
OLLO2.0_BV.ZIP |
Audio data
package 3/5: German speakers 21 - 30, Bavarian dialect (824 MB) |
| |
|
OLLO2.0_EP.ZIP |
Audio data
package 4/5: German speakers 31 - 40, Eastphalian dialect (905 MB) |
| |
|
OLLO2.0_FR.ZIP |
Audio data
package 5/5: French speakers 41 - 50 (1128 MB) |
| |
|
OLLO2.0_CORPUS
_SEGMENTATION.ZIP |
Two different
suggestions on how the OLLO corpus can be divided in train- and test
set (e.g. to evaluate the performance of different ASR systems by
using the same training and testlists). |
| |
|
OLLO2.0_LABELS.ZIP |
The OLLO
corpus was phonetically time-labeled, i.e., temporal positions of
phoneme boundaries have been determined automatically for each utterance,
making it suitable for tasks such as training of phoneme recognizers
(31 MB). |
| |
|
OLLO2.0_CALIBRATION.ZIP |
Calibration
sound files and normalization constants for all audio files, which
can be used to determine the original recording level (8 MB) |
| |
Literature:
"Oldenburg Logatome Speech Corpus (OLLO) for Speech Recognition
Experiments with Humans and Machines", Thorsten Wesker, Bernd
Meyer, Kirsten Wagener, Jörn Anemüller, Alfred Mertins, Birger
Kollmeier. Interspeech 2005. 4-8 September 2005, Lisbon, Portugal.
[pdf,
208 kB]
|
print