Page 34 - Spring 2015

Pushing the envelope of Auditory Research with Cochlear Implants
The Weird
There are many things that we do not understand about the auditory system and electrical stimulation. For example, most typical hearing and CI listeners cannot detect changes in temporal information in acoustic or electrical pulses be- yond 200-300 pulses/s, meaning that individuals cannot tell the difference between 400 and 600 pulses/s (van Hoesel, 2007; Kong et al., 2009). However, there is a small subset of CI listeners who can detect temporal information changes near 1,000 pulses/s (Kong and Carlyon, 2010; Noel and Ed- dington, 2013). These exceptional performers challenge our understanding of the temporal precision of the auditory nerve and neural encoding of electrical stimulation.
I described above how there is poor spectral resolution in CI users. A good engineering approach to solving this problem is to increase the spectral resolution with more electrodes and smaller electrical fields. Smaller electrical fields can be achieved by altering the placement of the ground electrode to other polarity configurations (i.e., change from monopolar to bipolar or tripolar stimulation). However, by doing this, even though we have increased spectral resolution and thus achieved our goal of presenting the CI user stimuli that are a little closer to typical acoustic hearing, these configurations produce mixed results at best and often poorer speech un- derstanding in CI users (Pfingst et al., 2001). In other words, exquisite spectral resolution is a hallmark of typical hearing; however, better spectral resolution in a CI produces mostly no change or poorer speech understanding. This may be a result of us just not understanding enough about how hear- ing works. Or could it be a result of the necessity to excite something in the cochlea? Again, we have an impaired ear and we need to convey the information somehow. But bet- ter spectral resolution may increase the probability that the information is not conveyed well because of portions of the cochlea that are lacking in ganglia and neurons to transmit information (i.e., neural dead regions; Shannon et al., 2001). Therefore, the engineer’s approach to the problem may be inappropriate. Having large current spreads and bandwidths will ensure that at least some neurons are transmitting the information, even if that is not how we think about how speech is encoding with typical hearing.
The Vocoder as a CI simulation:
To Model electrical stimulation and
To Improve our Understanding of It
by Reducing Variability
Ever since Shannon et al. (1995), the vocoder has produced a cottage industry of auditory research. The appeal of the vocoder is that it essentially works like the vocoder-centric speech processing shown in Figure 2. The only difference in the processing is that instead of using modulated electrical pulse trains to encode envelopes, sine tones or narrowband noises are used. So the goal of many vocoder studies is to simulate some aspect of CI processing, and the major ad- vantage of this technique is that normal-hearing listeners, for the most part, lack the variability of the real CI users. Us- ing such an approach, the high-performing CI users tend to match the normal-hearing listeners presented the vocoded signals (Friesen et al., 2001).
CI simulations are not limited to speech signals. Some peo- ple have used band-limited pulse trains to simulated electri- cal pulse trains, also with success in mirroring effects be- tween groups (Kan et al., 2013). However, one needs to take care. Acoustic signals can never replicate electrical stimula- tion because acoustical stimulation needs to follow certain physical laws and the electrical stimulation has a different set of rules. However, perfect replication is not the point of a vocoder. A good vocoder experiment targets a specific facet of CI processing or electrical stimulation to better under- stand the CI data. The simulation will never be perfect but has many merits and the error bars will be much more man- ageable in the normal-hearing listeners.
A great example is a paper by Oxenham and Kreft (2014) that modeled the current spread from electrodes to explain the masking of speech by different types of maskers. Their goal was to target a specific aspect of the CI with the simula- tion, which produced a congruence between the patterns of data across populations. On some levels, the debate about the worthiness of vocoder work as a CI simulation is one on the basic usefulness of models. Models are simplifica- tions of real life, miniature versions of larger more compli- cated things. In many cases, the best models are those that are simple, which is why the vocoder does a wonderful job of cutting through all of confounding factors that increase variability in real CI data.
32 | Acoustics Today | Spring 2015