Nonlinear spectrum compression for the hearing impaired via a frequency-domain processing algorithm

Abstract

In this paper, the results of both normal-hearing, and profoundly hearing-impaired adults, tested with spectrum compressed speech via the modified chirp-z algorithm, with and without visual stimuli, are reported. Ten normal-hearing adult listeners and five profoundly hearing-impaired adult listeners were asked to identify nonsense syllables presented auditorily and bimodally (audition and vision) via video tape in two conditions: lowpass filtered or unprocessed, and spectrum compressed. The lowpass filtered and spectrum compressed speech occupies the same spectrum width of 840 Hz; at 900 Hz and above, the attenuation is at least 60 dB. The spectrum compression is performed by means of a modified chirp-z algorithm, and is described in this paper. The testing results are significant and are reported in this paper. While the signal processing approach is somewhat intensive, the realtime throughput delay is small. Recent advances in hardware speed suggest that realization in a hearing aid is feasible.