Second Biennial Hearing Aid Research and Development Conference
September 22-24, 1997
National Institutes of Health
Predicting Subjective Sound Quality Ratings from Objective Measures
Catherine V. Palmer,
University of Pittsburgh
The two investigations described below suggest a method for transforming coherence values in order to predict sound quality judgments. Hearing aid distortion is most often measured according to the ANSI S3.22-1987 standard. When distortion is limited to the high frequency region or to increased input levels, the harmonic distortion measured at ANSI test frequencies and level may appear acceptable for a hearing aid that is nonetheless described by listeners as having poor sound quality. Recent reports have suggested measurement of high-frequency intermodulation distortion using coherence-function analysis which produces a number from 0 to 1 that indicates what portion of the output signal is due only to the input signal. Coherence functions currently are not widely measured because of the paucity of the data relating the measurements to subjective sound quality ratings.
Sound quality judgments of speech and music were obtained from 15 normally hearing and 11 hearing-impaired subjects on two binaural pairs of laboratory hearing aids with matched frequency/gain responses and differing only in coherence measurements as a function of input intensity. The set of hearing aids with higher coherence values as a function of input were judged to have superior sound quality. In order to compare the subjective findings with the physical measurement, coherence-function values were transformed by multiplying the 3 kHz value by 0.65, the highest average subjective sound quality rating. This produced good (+/- 25%) agreement between the subjective ratings and the transformed coherence values. The theoretical justification for the transformation was that limitations in bandwidth and response smoothness of the hearing aids used in this study appear to have limited the subjective quality ratings to a ceiling of 65%. The fact that a single-value multiplier appears adequate, and that its value is equal to the maximum value for the average quality ratings begged for replication and further investigation.
In a second investigation 15 normally hearing and 15 hearing-impaired individuals with no hearing aid experience judged the sound quality of three hearing aids matched for frequency/gain response and again differing only in coherence values across input levels. The subjects, signal judged (continuous discourse), signal level (70 and 90 dB SPL), and hearing aids were all different from the first experiment. The same coherence transformation was used employing a multiplication factor of .66 (the highest average subjective rating). Excellent agreement was found between the subjective ratings and transformed coherence-function values (+/-8%). Better agreement in the second study may be explained by the inclusion of previous hearing aid users in the first investigation.
The results support the use of coherence-function values applied to appropriate hearing aids in selecting amplification systems.
Return to Conference Abstracts List