Application of Real-Time Loudness Models Can Improve Speech Recognition for Cochlear Implant Users

摘要

The aim of cochlear implant (CI) stimulation strategies is to appropriately encode the important aspects of sound into a pattern of electrical stimulation. Recent research using numerical models of loudness perception has identified that there are large differences between how loudness is encoded by existing CI sound-processing strategies and how loudness is experienced by normally hearing listeners. In this paper, we present a new CI sound-coding algorithm aimed at addressing these discrepancies. This strategy, named SCORE, uses models of electric and acoustic loudness to modify the output of an existing CI sound-processing scheme in real time, so that the loudness changes are more accurately represented in the patterns of electrical stimulation. Five subjects (six implanted ears) were tested for understanding of speech presented at relatively low levels in quiet conditions. Using SCORE, subjects demonstrated an average 8.8 percentage-point statistically significant improvement $(p < 0.02)$ in the number of words correctly identified relative to ACE, a commonly used stimulation strategy. These findings show that loudness changes over time are important for speech intelligibility, and that improving loudness coding in existing CI devices may lead to perceptual benefits.