Re: Speech+Noise corpora

[Phil Green <p.green@xxxxxxxxxxxxxx>]

Hi Max,

>
>We are working on several new algorithms to enable us to extract speech
>signals in noisy environments, and are seeking ways to objectively validate
>the capabilities of our algorithms.  Does anyone on the list know of a set
>of well characterized speech and noise files that are available for
>downloading and that may have been used by others in evaluations of noise
>reduction strategies?  Examples of conditions that we would be interested
>in include:  speakers and stationary noise, various types of non-stationary
>noise, two overlapping speakers, speech and music, car environments,
>strongly echoic environments, etc.
>

The AURORA project (http://www.icp.inpg.fr/ELRA/aurora.html) is a focus for
robust ASR work at the moment. The material is digit strings from the TIDIGITS
database + various added noises at different levels.

Soon to be released is SpeechDat-Car, digit strings recorded in cars - will be
available through the ELRA site above.

On a smaller scale, you can get the 100 sound mixtures used in Martin Cooke and
Guy Brown's Ph.D.s from Martin's web site: http://www.dcs.shef.ac.uk/~martin/


*********************************
Professor Phil Green
Speech and Hearing Research Group
Department of Computer Science
University of Sheffield
Regent Court
211 Portobello St.,
Sheffield S1 4DP
UK

phone: (44) 114 22 21828
fax:   (44) 114 22 21810
email: p.green@dcs.shef.ac.uk
www:   http://www.dcs.shef.ac.uk/people/P.Green/
***********************************************