TY - JOUR
T1 - The intelligibility of speech with "holes" in the spectrum
AU - Kasturi, Kalyan
AU - Loizou, Philipos C.
AU - Dorman, Michael
AU - Spahr, Tony
PY - 2002/9
Y1 - 2002/9
N2 - The intelligibility of speech having either a single "hole" in various bands or having two "holes" in disjoint or adjacent bands in the spectrum was assessed with normal-hearing listeners. In experiment 1, the effect of spectral "holes" on vowel and consonant recognition was evaluated using speech processed through six frequency bands, and synthesized as a sum of sine waves. Results showed a modest decrease in vowel and consonant recognition performance when a single hole was introduced in the low- and high-frequency regions of the spectrum, respectively. When two spectral holes were introduced, vowel recognition was sensitive to the location of the holes, while consonant recognition remained constant around 70% correct, even when the middle- and high-frequency speech information was missing. The data from experiment 1 were used in experiment 2 to derive frequency-importance functions based on a least-squares approach. The shapes of the frequency-importance functions were found to be different for consonants and vowels in agreement with the notion that different cues are used by listeners to identify consonants and vowels. For vowels, there was unequal weighting across the various channels, while for consonants the frequency-importance function was relatively flat, suggesting that all bands contributed equally to consonant identification.
AB - The intelligibility of speech having either a single "hole" in various bands or having two "holes" in disjoint or adjacent bands in the spectrum was assessed with normal-hearing listeners. In experiment 1, the effect of spectral "holes" on vowel and consonant recognition was evaluated using speech processed through six frequency bands, and synthesized as a sum of sine waves. Results showed a modest decrease in vowel and consonant recognition performance when a single hole was introduced in the low- and high-frequency regions of the spectrum, respectively. When two spectral holes were introduced, vowel recognition was sensitive to the location of the holes, while consonant recognition remained constant around 70% correct, even when the middle- and high-frequency speech information was missing. The data from experiment 1 were used in experiment 2 to derive frequency-importance functions based on a least-squares approach. The shapes of the frequency-importance functions were found to be different for consonants and vowels in agreement with the notion that different cues are used by listeners to identify consonants and vowels. For vowels, there was unequal weighting across the various channels, while for consonants the frequency-importance function was relatively flat, suggesting that all bands contributed equally to consonant identification.
UR - http://www.scopus.com/inward/record.url?scp=0036713102&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=0036713102&partnerID=8YFLogxK
U2 - 10.1121/1.1498855
DO - 10.1121/1.1498855
M3 - Article
C2 - 12243158
AN - SCOPUS:0036713102
SN - 0001-4966
VL - 112
SP - 1102
EP - 1111
JO - Journal of the Acoustical Society of America
JF - Journal of the Acoustical Society of America
IS - 3 I
ER -