Distorting the spectral representation of speech signals detrimentally affects speech perception, acutely at least. Studies have examined different distortions including upward shifts in the frequency spectrum, linear and non-linear warping, and spectral compression/expansion. Often, subjects have been given little opportunity to adapt to the distorted speech signals. However, some studies have demonstrated that the detrimental effect of upward shifts in the frequency spectrum can be significantly reduced over three hours of training.
Shannon et al. (2002) examined a spectral distortion in connection with 'holes' or dead regions in hearing. Here, the spectral information that would normally be represented in the hole region was shifted to channels either side of the hole. Reassignment partially preserves the information that would be lost but also entails warping the place-code of the spectrum which may in itself lead to poorer performance. Results suggested rerouting information around a hole was no better than simply dropping it. This has implications for cochlear implant patients for whom the pathology of hearing loss may involve regions of dead/dysfunctional neurons (leading to 'holes' in the tonotopic representation of spectral information). However, in this study too, subjects received no training.
Here we examine the impact of a hole in the midfrequency region of the spectrum and the ability of listeners to adapt to spectral reassignments similar to those investigated by Shannon and colleagues. Noise-band speech processors were created with three output channels either side of the hole. Spectral information that would normally have been presented to the hole was either dropped or reassigned. Subjects trained in two different reassignment conditions. Performance in speech recognition improved considerably, in one condition sentence scores rising from 32% to 70% after three hours of training. Another group of subjects was trained in a condition where information from the hole region was simply dropped. Though some improvement was observed, scores at the end of training (e.g. 34% for sentences) were significantly lower than the first group achieved in the reassignment conditions. Although it is not possible to say whether further training would have resulted in complete adaptation in the reassignment conditions, these results are consistent with those from other studies that have demonstrated a significant reduction of the detrimental effects of distorting the spectral representation of speech with training.