The Articulation Index
and Hearing Aid Fitting:
The Bad News and the Good News
by Larry Revit, hearing scientist
Recently, our marketplace has expressed an interest in using the Articulation Index (AI) to assist in hearing aid fitting. Frye Electronics has been investigating efficacy of the AI in hearing aid fitting for over two-and-a-half years. Contained below is a summary of our findings, to date.
First the bad news, and then the good news.
The Bad News
The AI has not been validated as a hearing aid fitting method. Unlike the NAL, POGO, Berger, Lybarger, and Libby methods, no clinical study has shown the AI to be an effective fitting tool. To the contrary, even a very recent report1 shows the AI method to be either no better, or else poorer, than NAL or POGO in maximizing word-intelligibility scores. Keep in mind that almost any hearing aid will raise the AI score.
Maximizing the AI score can easily cause problems in a hearing aid fitting. In most cases, maximizing the AI is accomplished either by raising the overall gain -- something which hearing aid users don't generally tolerate -- or by increasing the high-frequency response --something which, in some cases, may have merit, but in others, may cause discomfort and feedback problems, especially with ski-slope losses.
The AI varies significantly for different listening tasks. The AI formula changes with every listening environment and set of speech-test materials. The AI's effectiveness in predicting speech intelligibility is inseparably tied to the particular, pre-selected listening conditions and materials accounted for in an AI formula. Thus, one hearing aid could have a high AI score and give good speech intelligibility for one listening condition, and the same hearing aid could have a high AI score and result in poor speech intelligibility for a different listening condition. Furthermore, acoustically different hearing-aids can have the same AI score. (which hearing aid is better???...)
Estimating the AI accurately is no trivial matter. Estimating the AI accurately requires an accurate estimate of aided sound-field thresholds. Two currently available "easy" AI methods use a potentially inaccurate procedure to estimate aided sound-field thresholds. Both METHODS call for estimating aided sound-field thresholds by adding measured insertion gain to earphone thresholds. One factor determining the insertion gain is the real-ear unaided response (REUR). Yet there is no evidence that REUR is correlated with earphone thresholds. So mixing insertion gain with earphone thresholds to predict aided sound-field thresholds introduces an error associated with an extraneous variable: the individual variability of the REUR.
But there is some Good News:
The accuracy of AI may soon improve. It may yet be possible that an accurate estimate of aided sound-field thresholds can be made through the appropriate real-ear probe measurements and correction factors. We are working on this, and we're confident that an improved METHOD can be devised.
The AI can be an important counseling and selling tool. There can be no better way to sell a hearing aid than to display to the client that he or she will understand speech better with a hearing aid than without one. So displaying an unaided-versus-aided improvement on the AI (a speech-intelligibility-related index) should help the dispenser convey to the client the attractiveness of buying a hearing aid. That will be good for everyone. And that is why Frye Electronics is working on developing an accurate AI method.
But we won't put it in our analyzers until we know it's accurate.
Footnotes
- C.M. Rankovic, "An application of the articulation index to hearing aid fitting." Journal of Speech and Hearing Research, Vol. 34, No. 2, pp. 391-402, April, 1991.
- H.G. Mueller, and M.C. Killion, "An easy for calculating the articulation index." The Hearing Journal, Vol. 43, No. 9, p. 14, September, 1990
- A contained within a portable hearing-aid analyzer manufactured in Canada.