Application of frequency importance functions to directivity for prediction of benefit in uniform fields

The purpose of the current investigation was to systematically examine two of the assumptions central to the application of Articulation Index weighted Directivity Index (AI-DI) to the prediction of directional benefit across three groups of listeners differentiated by degree and configuration of hearing loss. Specifically, the assumption that (1) changes in speech recognition performance are predictable from frequency specific changes in calculated audibility after applying directivity index (DI) values and (2) applying appropriate frequency importance functions would increase the accuracy of AI-DI predictions of directional benefit were evaluated.

The output of a single hearing aid for a speech in noise input was recorded to produce high and low directivity (directional and omnidirectional microphone modes) segments. These segments were then high-pass and low-pass filtered into low- and high-frequency regions and acoustically mixed to generate the eight frequency-specific directivity combinations. All recordings were made through an acoustic manikin placed in a single room, surrounded by five uncorrelated noise sources. The aided sentence recognition, in noise, for three groups of 12 adult participants with symmetrical sensorineural hearing impairment, was then measured across the eight listening conditions. The three groups were differentiated by degree and type of hearing loss including "sloping," "flat," and "severe" configurations. The frequency-specific DI values for each of the eight listening conditions were applied to the calculation of frequency specific noise levels. These corrected noise levels were then used to calculate an Articulation Index using the Speech Intelligibility Index (SII, ). These SII values were then compared with measured speech recognition under the same eight listening conditions. Directional benefit values were then calculated by subtracting the performance of individual participants on the Connected Speech Test (CST) in omnidirectional mode from performance in all other filter conditions. The changes in average DI and AI-DI (using three different frequency importance functions) that existed between omnidirectional and the other seven filter conditions were then calculated for comparison to directional benefit values.

The speech recognition data revealed a complex interaction between filter condition and group. Despite this interaction, highly significant positive correlations were found between participants' speech recognition scores and the corresponding SII calculation for all three hearing loss groups. Individual subjects' measured directional benefit was highly correlated with changes in DI. Similar correlations were found for average DI and all three AI-DI weighting methods.

As expected, performance and calculated SII values were in good agreement across conditions supporting the hypothesis that DI provides a reasonable frequency-specific estimate of signal-to-noise ratio changes in the test environment. The results further support the use of AI-DI or average DI for prediction of directional benefit. The choice of importance weighting across frequency (flat or frequency importance function based), however, did not improve the accuracy of these predictions; therefore, a simple average DI is advocated. Further, the prediction of absolute directional benefit across hearing loss groups from traditional AI-DI calculations may lead to error if the negative effects of hearing loss on speech understanding, and how these effects vary with degree of hearing loss, are not considered as a contributing factor.