Signal processing for cochlear implants and low-rate speech coding

Abstract

Summary form only given. Cochlear implants are now established as a new option for individuals with profound (sensorineural) hearing impairment. Many of the cochlear implant patients are able to understand speech without lip-reading, and some can communicate over the phone. The success of cochlear implants can be attributed to the combined efforts of scientists from various disciplines including bioengineering, physiology, and signal processing. Signal processing, in particular, played an important role in the development of various techniques for deriving electrical stimuli from the speech signal. Depending on the type of spectral information that was extracted from the acoustic signal, different speech processing strategies were developed over the years. The amount of spectral information that can be derived from the speech signal and delivered to the electrodes is limited, since the implant users have a small number (6-22) of electrodes. The designers of cochlear implants, much like the designers of speech coders, are therefore faced with the challenge of developing signal processing strategies that can extract a small, yet sufficient, amount of spectral information from the speech signal without compromising speech intelligibility and/or quality. Cochlear implants also provide us with a unique opportunity to study speech perception and investigate the perceptual limits of the auditory system. We can investigate, for example, the effect of limited spectral and intensity resolution on speech understanding, and ask questions such as "What is the smallest number of channels or what is the smallest number of discriminable intensity steps needed to understand speech?" The answers to such questions could potentially be used for the design of very low rate speech coders. This article provides an overview of various signal-processing techniques that have been used for cochlear prosthesis over the past 25 years, and also present some results from our intelligibility studies on the number of channels and quantization steps needed to understand speech.