Page 32 - Spring 2015

Pushing the envelope of Auditory Research with Cochlear Implants
Figure 4. Speech understanding for 8 late onset of deafness (typi- cally short duration of hearing loss and deafness) and 8 early on- set of deafness (typically long duration of hearing loss and deafness) adult CI listeners. All listeners received their CI as an adult. The 2 groups are matched in age, with an average age of 57 years. Speech understanding was measured using Az-Bio sentence in quiet or in a multitalker babble with +10-dB signal-to-noise ratio. Large circles are the average speech understanding ± 1 SD. Small circles show in- dividual speech understanding scores. Data are from a subset of data collected in the Goupell Laboratory at the University of Maryland, College Park.
In some ways, the engineer’s approach is one that assumes that more information is always better, and so the real limi- tations rely in better technology. Perhaps for this reason, there is good reason why the last major breakthrough was two decades ago. (There have been many advances in the past 20 years such as increased battery life, smaller and more durable devices, and microphone beamforming to better ac- commodate background noise; however, there has not been another revolutionary increase in speech understanding to what occurred in the 1990s.) In contrast, one could take a different approach that attempts to present only the most important information because there is a limited amount of information that a CI can convey. Such an idea is relatively less prevalent but exists in some speech-processing strate- gies. Would it not then be prudent to rethink how we pres- ent the information to maximize the amount of “useful” in- put the auditory system receives? For example, even though some CI users have 22 contacts on their arrays, one would think this would create 22 information channels. Sadly, this
does not happen. Because of a technological limitation of how the current travels through the cochlea (ultimately, where the ground electrode is located, which determines the path that the current takes), the electrical fields are quite broad, about 4-5 mm (Nelson et al., 2008). For a 35-mm cochlea, this means that each electrode excites about 1/8th of the cochlea. Therefore, it may not be surprising that CI users utilize only 7 or 8 independent information channels, even though they have 22 intracochlear electrodes (Friesen et al., 2001). Hopefully, someday soon we will provide more independent information channels through technological advances.
If CI users really only get information about eight channels, how do we present the information? As an alternative ap- proach, one could simply avoid trying to pass the full signals to the auditory system. One way to achieve this would be to only activate eight channels at a time, the ones with the high- est energy, because those channels have the most important information to convey (e.g., vowel formants). CIs that use peak-picking speech-processing strategies where only a subset of the possible electrodes fire are common. Such an idea is not dissimilar to removing information that might be masked in mp3 encoding. Also, along this line of thought of “less is more,” many investigators have shown that remov- ing electrodes that are “bad” for various reasons (e.g., bad modulation detection performance) can improve speech understanding (Garadat et al., 2012).
Yet another approach would be to do as much to the signal before it is passed to the auditory system. In other words, why not preprocess the signals and attempt to present rela- tively clean representations of the signals? Separating out noise before giving it to the auditory system has been an ef- fective approach in hearing aids for years and is a simple and elegant solution to our problem. Of course, if this were an easy task, we would be finished with CI research as a whole. Separating out targets from noise without a priori informa- tion is a nontrivial task. Therefore, moving CI performance to even higher levels will likely need to involve not only basic research and basic understanding of how the auditory sys- tem encodes electrical stimulation but also ways of maxi- mizing important information transfer (in contrast to forc- ing as much information into the auditory system as we can, we need to be judicious in our choice of information) and removing as much of the unwanted information like back- ground noise as we can before presenting information to the auditory system.
30 | Acoustics Today | Spring 2015