Comparison of Two Speech Processing Schemes Using Normal-hearing Subjects

Two speech processing schemes have been implemented on a real-time laboratory speech processor for speech perception studies on normal-hearing subjects. The first scheme presents the information of two spectral components, the first and second formant of the speech signal, and will be referred to as the ZC (zero-crossing) scheme. The second scheme presents the information of four spectral components, four spectral peaks of the speech signal determined from the output of a filter bank, and will be referred to as the FB (Filter-bank) scheme. Perceptual studies were conducted on 4 normal-hearing subjects. Two of the 4 subjects were first trained and tested using the ZC scheme while the other 2 used the FB scheme. The training and testing schedules were then repeated for the unused scheme. Vowel perception performance was similar across subjects and speech processing schemes. Percentage correct scores for consonant perception in noise were higher for the FB scheme than for the ZC scheme at signal-to-noise ratios ranging from 5 to 20 dB. Percentage correct scores for open set CNC words and speech tracking scores were also better for FB than for ZC.