8th Speech in Noise Workshop, 7-8 January 2016, Groningen

Real time hearing impairement simulator for speech intelligibility measurements

Nicolas Grimault(a)
Auditory Cognition and Psychoacoustic team, Centre de Recherche en Neurosciences de Lyon, CNRS, Université Lyon 1, FR

Etienne Parizet
Laboratoire Vibrations Acoustique, INSA-Lyon, FR

Aexandra Corneyllie
Auditory Cognition and Psychoacoustic team, Centre de Recherche en Neurosciences de Lyon, CNRS, Université Lyon 1, FR

Samuel Garcia
Auditory Cognition and Psychoacoustic team, Centre de Recherche en Neurosciences de Lyon, CNRS, Université Lyon 1, FR

Laurent Brocolini
Laboratoire Vibrations Acoustique, INSA-Lyon, FR

(a) Presenting
(b) Attending

The AIDA projects aims at improving speech intelligibility in cars for people suffering from mild hearing losses. Signal processing will be used to modify speech messages provided by a car (e.g. from the navigation system), in order to make them more intelligible to the driver.

This will necessitate many experiments using hearing-impaired people. In order to ease such experiments, it was decided to use a hearing-loss simulator. This simulator will make it possible to evaluate the effectiveness of some signal processing techniques using normal hearing people, while simulating a given threshold profile.

The simulator reproduces the enlargement of the auditory filters as well as the loss of recruitment at low level of presentation. The procedure is basically an inverse gammachirp auditory filter bank as first described by Irino. The algorithm is implemented in Python and parallelized using the graphic card resources. Such a design enables the algorithm to work in real time, so that the signal produced by a software conducting the intelligibility experiment is modified before being presented to the listener.

This paper will present the simulator as well as an experiment conducted to evaluate its accuracy. First of all, Speech Reception Threshold has been measured in three typical car interior noises, using a french version of the Four Alternative Auditory Feature test (Foster and Haggard, 2005). Participants were 76 people, recruited either as customers of an audioprothesist, either according to age criterion (over 55). Tonal audiometry allowed to separate participants in three groups, from normal hearing to a mild hearing loss. SRT was measured within these three groups.

Then, twenty young normal-hearing subjects were recruited. The simulator allowed to recreate the hearing ability of a typical member of the mild hearing loss group, as defined in the first step. The SRT was measured and compared to the one obtained with the previous subjects.

This device allows an easier recruitement of subjects, for which the effectiveness of speech emphasizing techniques will be evaluated.

Last modified 2016-05-12 14:22:09