SS07: PSYCHOACOUSTICS, SPEECH PERCEPTION AND AUDITORY SIGNAL PROCESSING

MODELING BINAURAL SPEECH INTELLIGIBILITY OF HEARING-IMPAIRED LISTENERS IN ADVERSE CONDITIONS

Beutelmann R, Brand T, Kollmeier B

Medizinische Physik, Universität Oldenburg, Germany

Acoustically unfavourable situations (characterized by several sound sources and reverberation) are especially challenging for  , hearing-impaired listeners. One of the essential aids in adverse conditions is binaural hearing, but the binaural processing of hearing-impaired listeners is influenced by their hearing loss, due to distorted input from the periphery and/or because of deficiencies on higher levels of the auditory system. This talk presents binaural speech intelligibility measurements and model simulations with hearing-impaired subjects having various degrees of hearing loss. Speech reception thresholds (SRT) in noise were measured in a number of situations incorporating a speech-shaped stationary or fluctuating noise source at various azimuth angles and in different room acoustical conditions. In addition to their overall decreased SRT by up to 10 dB SNR compared to a normally-hearing control group, the subjects suffered from a loss of binaural unmasking in spatially favorable conditions, mostly between 5 and 10 dB SNR, sometimes up to 20 dB SNR, depending on condition and subject. The observed data were compared to predictions of a binaural speech intelligibility model (Beutelmann and Brand 2006, JASA), a multi-channel equalization-cancellation process used as a binaural front-end for the Speech Intelligibility Index. The hearing losses were included in form of interaurally uncorrelated internal masking noises spectrally shaped according to the individual audiograms. The combined effect of audibility and of interaural decorrelation on binaural speech intelligibility can be predicted very well (correlation coefficients between predictions and observed data above 0.9). Overall, about 70% of the variance due to hearing impairment can be predicted based on the audiogram. Suprathreshold and cognitive effects, which may be underlying the remaining variance, will require model extensions and further individual measures of the hearing-impairment.