Speech recognition in single- and multichannel cochlear implants

Two speech processing systems, a four-channel vocoder-based model and a single-channel model, were examined in a single patient. Five isolated, computer-generated vowels were presented in a closed-set paradigm and responses were plotted on a confusion matrix. With four-channel stimulation the overall score was 64% correct. The subject never confused vowels having a low-frequency second formant (F2) with high-frequency F2 items. In the single-channel configuration, the subject did not distinguish between low F2 and high F2 items and scored 29% overall. Consonant confusion matrices were generated for seven naturally produced VCV syllables. In the four-channel configuration the overall score was 70% correct. The nasal distinction was not made and some voicing errors were observed. In the single-channel case, the subject scored 27% correct. The stop-continuant distinction was not routinely made and most errors involved multiple distinctive features. Analysis of the processed speech items explain some of the confusions observed in the four-channel mode. [Work supported by NIH Grant NS-11804.]