Discrimination of speech processed by low-pass filtering and pitch-invariant frequency lowering

Consonant discriminability by normal-hearing listeners was studied for monosyllables processed using frequency lowering with either uniform (linear) or nonuniform (warped) compression of the frequency axis. The bandwidth of the compressed signal was either 2500 or 1250 Hz and five compression characteristics were studied, in addition to low-pass filtering, for each of the two bandwidths. Stimuli were 24 consonant-vowel syllables made up of 24 consonants and three vowels, recorded three times by each of two male and female speakers. A "roving-speaker, roving-vowel, roving-utterance," two-interval, forced-choiced procedure with feedback was employed. As expected, performance for the 2500-Hz bandwidth was superior to that for the 1250-Hz bandwidth. For each bandwidth, compression schemes which accomplished greater lowering of the high frequencies relative to the low frequencies resulted in the highest performance. Overall performance for the best frequency-lowering system studied, however, was roughly comparable to that obtained for low-pass filtering. Patterns of performance over articulatory features were different for frequency lowering and filtering. In general, lowering was superior to filtering for contrasts of fricative sounds but inferior for contrasts of nasals and semivowels.