CHR AJPI Feb 2006

aids over 100 years ago, the search has been on for a way to fully compensate for all aspects of hearing impairment by hearing aid amplification. To date, this goal has not been achieved. However, as hearing aid technology has developed from the era of carbon aids to the present-day digital signal processors, much progress has been made in making optimal use of the available technology. The vast majority of improvements in amplification strategies have been achieved by trial and error using empirical testing; the physiological basis for optimal hearing aid amplification is generally not well understood. Consider the development of prescriptions for linear amplification. Early studies suggested “mirroring of the audiogram,” such that 1 dB of gain is applied for each 1 dB of hearing loss at each frequency. While this appeared to be satisfactory for conductive hearing losses, it was found to amplify loud sounds excessively for sensorineural losses. Consequently, it was proposed that a constant could be subtracted from the gain to bring it down to an acceptable level. Watson and Knudsen (1940) refined this idea by suggesting that the gain should be reduced additionally by a factor depending on most comfortable equal loudness curves. Shortly thereafter, Lybarger (1944) suggested the alternative “half gain rule” that formed the basis for many subsequent linear amplification prescription schemes. The goal of these linear prescriptions is to optimise audibility, comfort, and speech intelligibility; the differences between the various prescription schemes reflect different weightings of the importance of these desired perceptual outcomes. But can we determine the physiological basis for a linear gain prescription? Bondy et al. (2004) used a computational model of the auditory periphery to determine the linear gain-frequency responses that would optimally restore normal average levels of auditory nerve activity for a range of audiograms. The model predictions very closely match the NAL-R prescription (Byrne and Dillon, 1986). This result indicates that in applying amplification that gives more normal auditory nerve activity on average, an optimal mix of audibility, comfort, and speech intelligibility is obtained for speech on average. We then attempted to extend these results to multiband compression on a phoneme-byphoneme basis, rather than just for speech on average. It is known from empirical studies that by adjusting the compression characteristics, it is possible to avoid distorted and uncomfortably loud signals, to reduce the intensity differences between phonemes or syllables, to provide automatic volume control, to increase sound comfort, to normalize loudness, to maximize intelligibility, or to reduce background noise (Dillon, 2001). However, the required compression parameters vary substantially among these goals; consequently, any one compression scheme tends to provide benefit in some but not all aspects of compensating for hearing impairment. Consistent with the empirical observations, using the auditory periphery model it was not possible to find one set of compression parameters that normalized the auditory