Real-Time Contrast Enhancement to Improve Speech Recognition
(a) and (b) Spectrograms of the unenhanced and enhanced VCV /aga/ as spoken by an adult male talker. Vertical lines in the spectrograms correspond to time windows used in (c) and (d), which show spectra for the unenhanced (thick-solid line) and enhanced (thin-dotted line) signals from 16-ms time segments centered at 135 and 152 ms, respectively. (e) Spectral smearing of the unenhanced and enhanced time segments in (c) with a moderate degree of smearing (thick-solid and thin-dotted lines, respectively). (f) Formant peak locations of the stimuli shown in (a) and (b) as derived from linear predictive coding.