Speaker under Stress Database
A Speech Database from 20/20 Speech
Voice control of computers is no longer the preserve of benign office
environments. With this liberation of the technology away from the
desktop has come a whole new set of challenges, including the
requirement to accurately recognise stressed, shouted and even
whispered speech produced under difficult acoustic conditions.
Most of the databases available today are recorded under acoustically
quiet conditions, with the speaker under no physiological or
psychological load. This data, although well matched to benign
environments, fails to provide an accurate picture of the variation of
the speech signal under different �real world' conditions.
Collected by 20/20 Speech Limited, the Speaker under Stress database
consists of over 9 hrs of recorded speech material, covering a range
of different applications including digits, command phrases and
vehicle registration plates, under a range of different speaker
conditions including physical and mental stress.
Part A: 16 Speakers (all male), 7hrs speech, digits, dictated vehicle
registration plates, control words.
Data was designed to cover three classes of variation:
Acoustic Environment - including reverberation, gunfire, interfering
speech and engine noise
Physiological Stress - include crawling, jogging and recovery Vocal
Style - including shouting and whispering
Data was recorded on a high quality headset-mounted microphone.
Part B: 15 Speakers (11 male and 4 female), 2 1/4 hrs speech, dictated
vehicle registration plates. Recorded in a sound proof booth, using a
high quality headset-mounted microphone, under controlled levels of
cognitive-workload stress. After each individual test, the subject was
asked to complete a NASA-TLX subjective workload questionnaire to
provide a subjective measure of the stress.
Pricing available on application
For further information please contact 20/20 Speech
email: firstname.lastname@example.org Tel:+44 (0)1684 585107
Just one of the family
The Speaker under Stress Database is just one of a family of quality
databases recorded by 20/20 Speech. These include the �Accents of the
British Isles Database' and the �Device Control Database'. Further
information regarding these databases and how they can used, can be
found on our website, www.aurix.com.If you have a specific requirement
for a database, 20/20 Speech also offer abespoke database service,
managing all aspects of project management, database design, recording
About 20/20 Speech
20/20 Speech Ltd has been at the forefront of developments in speech
technology for over 30 years, initially as the UK Government's Speech
Research Unit (SRU) and more recently as speech technology company
20/20 Speech Ltd (a joint venture between QinetiQ and NXT plc).
Through the innovative Aurix� range of products, solutions and
consultancy, 20/20 Speech offers a unique set of tools that enable
third parties to build speech driven interfaces for their
applications. Please visit www.aurix.com for further details.
20/20 Speech Limited Malvern Hills Science Park, Malvern,
Worcestershire, UK WR14 3SZTel: +44 (0)1684 585100 Web:
urix and 20/20 Speech are registered trademarks