FEATURE ARTICLE
The Design and Function of Cochlear Implants
Fusing medicine, neural science and engineering, these devices transform human speech into an electrical code that deafened ears can understand
Michael Dorman, Blake Wilson
Slicing the Spectrum
How many channels, these slices of the frequency spectrum, are
necessary to restore speech understanding? Following the early work
of Robert Shannon and others at the House Ear Institute in Los
Angeles, one of us (Dorman) answered this question with collaborator
Philip Loizou of the University of Texas at Dallas. We used
so–called bandpass filters to divide the spectrum of speech
into a relatively small number of frequency bands or channels. A
microprocessor measured the energy in each band every few
milliseconds and transformed the signal into
amplitude–modulated sine waves, each centered on one of the
frequency bands. When we played these simplified audio signals to
normal–hearing listeners, they understood 90 percent of the
words in simple sentences, even when we used as few as four
channels. Eight channels allowed them to identify 90 percent of
isolated words. Under noisy conditions, more channels were needed to
match this performance, and the more channels used, the better the
comprehension. These observations show that in a quiet environment,
speech can be well understood with a relatively small number of
channels—a fact that is central to the success of cochlear implants.
The variability of normal speech also helps. As our brain decodes
speech sounds, it uses cues, called formants, which
identify consonants and vowels by specific concentrations of energy
in the frequency spectrum. But formants do not have a fixed
frequency, even for the same sound, because vocal–tract
geometry varies from speaker to speaker. So instead of being
discrete points, the acoustic signatures of speech sounds can be
thought of as ellipses in frequency space. Even in infants, the
system that perceives speech is designed to be flexible so that it
can "hear through" variations in the signal to identity a
consonant or vowel. This flexibility allows a very reduced
description of speech to be recognized with accuracy.
» Post Comment