Two steady-state noise-excited vowel sounds of 200-ms duration were synthesized: one similar to a naturally produced whispered vowel, the other with energy removed outside a nominal 50-Hz band centered at each formant peak. The threshold for a brief sinusoidal probe tone was measured at frequencies corresponding to the peaks and troughs in the vowel spectra, in both forward and simultaneous masking. The masking functions for a flat-spectrum masker at each frequency were also measured, so that the vowel masking patterns could be expressed in terms of the equivalent broadband masker level. In the first experiment, masking patterns for the two maskers were measured with three normal-hearing subjects. Increasing spectral contrast was effective in increasing peak–valley differences in the masking patterns up to 2.5 kHz. Contrast was greater in forward than in simultaneous masking, suggesting that the internal representation of a vowel may be enhanced by suppression. In a second experiment, masking patterns and flat-spectrum masking functions were measured with three listeners with moderate sensorineural hearing impairments. The enhancement produced by increasing spectral contrast was less than that obtained with normal-hearing listeners, but still present up to 2.5 kHz. A comparison of forward- and simultaneous-masking patterns again revealed an increase in peak–valley differences, suggesting that suppression mechanisms were still effective in these listeners. The ability to enhance the internal representation of speechlike sounds in hearing-impaired listeners may have implications for the design of signal-processing hearing aids.