Dear Dick, Chuping, and auditory.org list:
Thank you (Dick) for your summary of Brian Moore?s talk. I wasn?t present, but I am familiar with Brian?s ideas on this topic.
Contrary to Brian?s assertion that the audiogram isn?t enough, I have had considerable success predicting speech intelligibility scores for hearing-impaired listeners using just the audiogram. I use Harvey Fletcher?s full-blown Articulation Index [H. Fletcher and R.H. Galt (1950), J. Acous. Soc. Am. 22, 89-151]. To use the AI, I model the hearing loss as noise, and calculate expected masking spread for any external noise, and then plug these into an otherwise unaltered Fletcher calculation.
I have published three evaluations of prediction accuracy of Fletcher?s AI for hearing-impaired listeners. These included quiet and noisy conditions (steady-state noise, not fluctuating), amplified and filtered speech, and a variety of audiograms [C.M. Rankovic (1997), Chapter 26 in Modeling Sensorineural Hearing Loss, edited by W. Jesteadt, Erlbaum, Mahwah, NJ; C.M. Rankovic (1998), J. Acous. Soc. Am. 103, 1043-1057]. I also evaluated Brian?s dead region hypothesis, using Brian?s own published data, and I was able to predict his speech intelligibility scores with very good accuracy [C.M. Rankovic (2002), J. Acous. Soc. Am. 111, 2545-2548].
I?ve recently extended Fletcher?s AI calculation to predict binaural speech intelligibility for normal and hearing-impaired listeners. This is necessary because, after all, most hearing aid fittings are binaural.
I believe there are important implications of my success with Fletcher?s articulation index for hearing-impaired listeners:
1. Fletcher?s calculation is the
culmination of decades of basic research at Bell Labs that investigated the
normal processes of hearing [Fletcher, H. (1995), edited by J.B. Allen, The ASA
Edition of Speech and Hearing in Communication available from http://asa.aip.org, C.M. Rankovic and J.B.
Allen (2000), Study of Speech and Hearing at Bell Telephone Laboratories, CD
available from http://asa.aip.org]. As
such, it includes extensive psychoacoustical modeling that considers intensity
level effects, loudness growth, masking spread, and more. Fletcher?s rigorous and elaborate
efforts led to a robust model for normal listeners to which extensions for
hearing loss can be added.
2. There is no denying that there are many psychoacoustical abnormalities associated with hearing loss--such as abnormal temporal processing mentioned above. Researchers (including myself) have struggled for decades to quantify the contribution of these factors to speech intelligibility. Nevertheless, it seems that the AUDIOGRAM encompasses these factors in a way that is sufficient for predicting speech intelligibility.
Within the next couple of months, I will make available a simple-to-use version of my AI calculation. It will feature audiogram entry and binaural capability. Please contact me if you are interested and I will keep you posted.
Christine Rankovic, PhD