[Consonant sounds processing in 22 channel cochlear implant device and consonant recognition of patients]

This report describes studies on electrode selection of the speech processor of a 22 channel cochlear implant device for 14 Japanese consonants, and on consonant recognition of 23 cochlear implant patients as they relate to electrode selection. The means of electrode selection was studied using a special device capable of detecting code signals delivered from the speech processor for electrode selection. Young male voices consisting of/a/+consonant+/a/ were presented as sound stimuli, and the electrode numbers selected were recorded for each consonant. The recorded numbers were grossly divided into two groups; a higher frequency range of about 500Hz to 4,000Hz and a lower range of 0Hz to about 1,100Hz. They showed characteristic patterns for each consonant, and these electrode number patterns were compared with patterns estimated from a sonogram. Although there were some differences between these two patterns in unvoiced stop consonants, they were roughly parallel to each other in other consonants. The information transfer ratio was calculated based on confusion matrices obtained from test results on 23 patients for 14 consonants. The ratio was high in voicing and semivowels, whereas it was relatively lower in plosives. This seems reasonable on the basis of patterns of electrode number transitions. Nasals and fricatives, however, showed lower information transfer ratios despite adequate electrode selection, and this may due to poorer information transition or the high frequency components of these consonants.