THE EFFECTS OF SYLLABIC COMPRESSION AND FREQUENCY SHAPING ON SPEECH-INTELLIGIBILITY IN HEARING-IMPAIRED PEOPLE

The effect of syllabic compression on speech intelligibility is rarely positive and in those cases that positive effects have been found, th e same positive results could in general be obtained by frequency shap ing of the frequency response curve. We programmed a syllabic compress or on a digital processor; the compressor differed from a conventional syllabic compressor by incorporating a delay in the signal path to su ppress overshoots and thus minimize transient distortion. Furthermore, the time constants were short: attack time of 5 msec and release time of 15 msec. The compressor was only active in the high-frequency band . An essentially linear signal was added to deliver the low-frequency speech components. The processing resulted in a frequency response tha t mirrored the hearing loss near threshold and became much flatter for higher level input signals. Speech intelligibility scores for nonsens e consonant-vowel-consonant words embedded in carrier phrases were det ermined for hearing-impaired persons with sloping audiograms and discr imination losses for speech. Results showed little additional effect o f frequency shaping to the existing improved speech score for compress ed speech. Optimum results were found for a compression ratio 2 with l ower speech scores for linear amplification and for compression ratio 8. We next determined the effect of providing high-frequency emphasis to the speech signal and/or to the compression control signal to compe nsate for the upward spread of masking. The frequency response at the root-mean-square level was adjusted according to the half-gain rule. T he positive effects of moderate compression could be found again; the high-frequency emphasis, however, was positive for the vowels but made consonant recognition poorer. We concluded that smoothing the speech intelligibility score improved for moderate compression with relative little effect of frequency shaping. Adding high-frequency emphasis to a half-gain rule response curve was not advantageous.