Discrimination of multichannel-compressed speech in noise: long-term learning in hearing-impaired subjects.

OBJECTIVE

It takes time for an individual to obtain optimal benefit from a new hearing aid. This research examines the possibility that similar long-term learning can be seen in consecutive laboratory studies of multichannel compression (MCC).

DESIGN

Three studies of different parameters of MCC processing, carried out over the period of 1 yr, included the same 15 hearing-impaired subjects and one identical MCC-processing condition. The full-range MCC had 8, 12, or 16 independent frequency channels, using a Robinson-Huntington compression algorithm. The City University of New York nonsense syllable test was modified to facilitate digital signal processing and control of the experiments. The subjects discriminated nonsense syllables (a female and a male voice) in speech spectrum noise at -5 to 15 dB signal-to-noise ratios (S/N). Conditions were not ideal for learning: subjects' experience with MCC-processed speech was limited to the laboratory and no trial-by-trial feedback was provided. Percent correct syllable discrimination and consonant confusion matrices were compared across experiments to observe the subjects' learning to listen with MCC processing.

RESULTS

All subjects combined, and 14 of 15 individual subjects, showed significant improvement across experiments. For the subject showing the maximum learning, the percent correct difference between the first and third experiments was equivalent to a 9.8 dB increase in S/N. The average learning for all subjects was equivalent to +3 dB S/N. The difference between the consonant-confusion matrices for the first and third experiments indicated that improved discrimination occurred for both manner and place information. The pattern of changes in the confusion matrices was consistent with improved use of the high-frequency information supplied by the MCC signal processing. A brief comparison of the results of the first experiment with a fourth experiment indicated that the learning was specific to MCC processing because it did not generalize to frequency-shaped linear amplification which was also studied in those two experiments.

CONCLUSIONS

These results indicate that specific long-term learning occurred for hearing-impaired subjects listening to nonsense syllables in noise with 8- to 16-channel MCC processing. Since previous experiments have provided subjects with much less listening experience, the results suggest that MCC with large numbers of channels may be much more beneficial for the hearing-impaired individuals than the results of previous experiments had indicated.