The question under consideration in this project was how to generate noise samples appropriate for reliably testing the quality of in-car speech recognition systems. Speech-driven applications are becoming more common in cars and it is important to be able to test how well such systems perform and to identify those aspects of the noise which may cause difficulties for the systems. This requires characterisation of the noise environment within a car. Such characterisations must then be used to efficiently compress the information so that testing can be performed economically.

Problem presented by
Dave Fish (J-Omega)

Study Group contributors
David Allwright (Smith Institute)
Melvin Brown (Smith Institute)
Richard Eyres (University of Bristol)
Tom Griffin (University of Bristol)
Sam Howison (University of Oxford)
Ralf Jacobs (University of Strathclyde)
Colin Please (University of Southampton)
Richard Rosing (University of Lancaster)
Eddie Wilson (University of Bristol)