Spectral enhancement to improve the intelligibility of speech in noise for hearing-impaired listeners.

At speech-to-noise ratios between -3 and 6 dB, many hearing-impaired listeners have difficulty in understanding speech, but spectrograms reveal that the formant peaks of voiced speech and some of the spectral peaks associated with unvoiced speech stand out against the background noise. Our speech-enhancement process is based on the assumption that increasing spectral contrast will result in improved intelligibility. The enhancement involves calculating an auditory excitation pattern from the magnitude spectrum of overlapping short segments of the speech signal. This pattern is convolved with a difference-of-Gaussians function whose bandwidth varies with frequency in the same way as the auditory filter bandwidth. Magnitude values from this enhanced pattern are combined with the unchanged phase spectrum from the original signal to produce the enhanced speech. The processing was used to enhance Boothroyd and Bench-Kowal-Bamford Audiometric lists which had been digitally combined with speech-shaped noise at speech-to-noise ratios between -3 and 6 dB. The subjects had moderate to severe sensorineural hearing losses. The processing produced small but significant improvements in intelligibility for the hearing-impaired listeners tested. Possibilities for improving the processing are discussed.