Speech recognition of hearing-impaired listeners: predictions from audibility and the limited role of high-frequency amplification.

Abstract

Two experiments were conducted to examine the relationship between audibility and speech recognition for individuals with sensorineural hearing losses ranging from mild to profound degrees. Speech scores measured using filtered sentences were compared to predictions based on the Speech Intelligibility Index (SII). The SII greatly overpredicted performance at high sensation levels, and for many listeners, it underpredicted performance at low sensation levels. To improve predictive accuracy, the SII needed to be modified. Scaling the index by a multiplicative proficiency factor was found to be inappropriate, and alternative modifications were explored. The data were best fitted using a method that combined the standard level distortion factor (which accounted for decrease in speech intelligibility at high presentation levels based on measurements of normal-hearing people) with individual frequency-dependent proficiency. This method was evaluated using broadband sentences and nonsense syllables tests. Results indicate that audibility cannot adequately explain speech recognition of many hearing-impaired listeners. Considerable variations from audibility-based predictions remained, especially for people with severe losses listening at high sensation levels. The data suggest that, contrary to the basis of the SII, information contained in each frequency band is not strictly additive. The data also suggest that for people with severe or profound losses at the high frequencies, amplification should only achieve a low or zero sensation level at this region, contrary to the implications of the unmodified SII.