Margaret Mortz
Dept. of Eng., Univ. of Denver, Denver, CO 80208
Christopher Schweitzer
Mark Terry
AudioLogic, Inc., Boulder, CO 80301
This research attempts to address the problem of hearing in noise by hearing-impaired listeners with digital algorithms motivated by the results of Carrell and Opie, 1992, who showed that temporal amplitude modulations helped normal-hearing listeners form auditory objects when exposed to time-varying sinusoidal sentences. The current processing algorithms involved segmenting speech materials into spectral bands and extracting a signed envelope from each band. The envelope in each band was then normalized and either directly added back to the normalized signal or ``edge enhanced'' with a derivative operator before recombination with the normalized signal. A compression algorithm using a hyperbolic tangent was used to compress the enhanced speech before renormalizing to the same power (rms) as the original speech. Processing variations included applying coherence rules across bands to ascertain when it was ``appropriate'' to modulate the signal with this technique. Standardized (BKB) sentences were scored for seven hearing-impaired listeners with and without the modulation processing. As a further control, preemphasized treble-enhanced samples were also compared. The same materials were used for subjective judgments on scaled perceptual difference measures, again comparing processed versus the unprocessed speech samples. Both types of measures produced encouraging results.