Speech intelligibility with a cochlear implant sound-processing scheme based on loudness models

A new sound processing scheme for cochlear implants, called SpeL, has been developed and evaluated. Unlike existing or previous schemes, SpeL is based on real-time numerical estimations of loudness. Computational models are used to estimate the loudness of incoming sounds, and to generate appropriate patterns of electrical stimulation to provide loudness characteristics for implant users that are similar to those of normally hearing listeners. The perceptual performance of an initial version of the SpeL scheme has been compared to that of the Speak (or ACE) scheme using tests of word recognition with 5 subjects. Preliminary results show that SpeL improved consonant recognition, at least at a relatively low speech level, but lowered vowel recognition, especially at a higher speech level. An overall increase in phoneme recognition with SpeL at a speech level of 50 dBA was found to be statistically significant. The significant decrease in scores observed at a level of 60 dBA is probably related mainly to the way acoustic frequencies were distributed across the available electrode positions in this version of SpeL.