Perceptual based speech enhancement for normal-hearing and hearing-impaired individuals

This paper describes a new noise suppression scheme with the goal of improving speech-in-noise perception for hearingimpaired listeners. Following the work of Tsoukalas et al. (1997) [4], Arehart et al (2003) [3] implemented and evaluated a noise suppression algorithm based on an approach that used the auditory masked threshold in conjunction with a version of spectral subtraction to adjust the enhancement parameters based on the masked threshold of the noise across the frequency spectrum. That original formulation was based on masking properties of the normal auditory system, with its theoretical underpinnings based on MPEG-4 audio coding [6]. We describe here a revised formulation, which is more suitable for hearing aid applications and which addresses changes in masking that occur with cochlear hearing loss. In contrast to previous formulations, the algorithm described here is implemented with generalized minimum mean square error estimators, which provide improvements over spectral subtraction estimators [1]. Second, the frequency resolution of the cochlea is described with auditory filter equivalent rectangular bandwidths (ERBs) [2] rather than the critical band scale. Third, estimation of the auditory masked thresholds and masking spreading functions are adjusted to address elevated thresholds and broader auditory filters characteristic of cochlear hearing loss. Fourth, the current algorithm does not include the tonality offset developed for use in MPEG-4 audio coding applications. The scheme also shows an overall improvement of 11% in the Itakura-Saito distortion measure.