Page 57 - Spring2019

important for recognition. Instead, the depicted strategy, continuous interleaved sampling (CIS; Wilson et al., 1991), makes no assumptions about how speech is produced or per- ceived and simply strives to represent the input in a way that will utilize most or all of the perceptual ranges of electrically evoked hearing as clearly as possible.
As shown, the strategy includes multiple channels of sound processing whose outputs are directed to the different elec- trodes in an array of electrodes implanted in the scala tym- pani (ST), one of three fluid-filled chambers along the length of the cochlea (see X-ray inset in Figure 1, which shows an electrode array in the ST). The channels differ only in the frequency range for the band-pass filter. The channel out- puts with high center frequencies for the filters are directed to electrodes at the basal end of the cochlea, which is most sensitive to high-frequency sounds in normal hearing (the tonotopic organization mentioned in A Snapshot of the His- tory), and the channel outputs with lower center frequencies are directed to electrodes toward the other (apical) end of the cochlea, which in normal hearing is most sensitive to sounds at lower frequencies.
The span of the frequencies across the band-pass filters typi- cally is from 300 Hz or lower to 6 kHz or higher, and the dis- tribution of frequencies is logarithmic, like the distribution of frequency sensitivities along the length of the cochlea in normal hearing. In each channel, the varying energy in the band-pass filter is sensed with an envelope detector, and then the output of the detector is “mapped” onto the narrow dy- namic range of electrically evoked hearing (5-20 dB for puls- es vs. 90 dB or more for normal hearing) using a logarithmic or power-law transformation. The envelope detector can be as simple as a low-pass filter followed by a rectifier (full wave or half wave) or as complex as the envelope output of a Hil- bert Transform. Both are effective. The compressed envelope signal from the nonlinear mapping function modulates a carrier of balanced biphasic pulses for each of the channels to represent the energy variations in the input. Those modu- lated pulse trains are directed to the intracochlear electrodes as previously described. Implant users are sensitive to both place of stimulation in the cochlea or auditory nerve and the rate or frequency of stimulation at each place (Simmons et al., 1965).
Present-day implants include 12-24 intracochlear electrodes; some users can rank all of their electrodes according to pitch, and most users can rank at least a substantial subset of the electrodes when the electrodes are stimulated separately and one at a time. (Note, however, that no more than eight elec-
Figure 2. Results from initial comparisons of the compressed analog (CA) and CIS processing strategies. Green lines, scores for subjects selected for their exceptionally high levels of per- formance with the CA strategy; blue lines, scores for subjects selected for their more typical levels of performance with that strategy. The tests included recognition of two-syllable words (Spondee); the Central Institute for the Deaf (CID) everyday sentences; sentences from the Speech-in-Noise test (SPIN) but here without the added noise; and the Northwestern University list six of monosyllabic words (NU-6). From Wilson and Dor- man (2018a), with permission.
trodes may be effective in a multichannel context, at least for ST implants and the current processing strategies; see Wilson and Dorman, 2008.) Also, users typically perceive increases in pitch with increases in the rate or frequency of stimula- tion, or the frequency of modulation for modulated pulse trains, at each electrode up to about 300 pulses/s or 300 Hz but with no increases in pitch with further increases in rate or frequency (e.g., Zeng, 2002). For that reason, the cutoff of the low-pass filter in each of the processing channels usually is set at 200-400 Hz to include most or all of the range over which different frequencies in the modulation waveforms can be perceived as different pitches. Fortuitously, the 400- Hz choice also includes the full range of the fundamental frequencies in voiced speech for men, women, and children. The pulse rate for each channel is the same across channels and is usually set at four times the cutoff frequencies (which also are uniform across channels) to minimize ambiguities in the perception of the envelope (modulation) signals that can occur at lower rates (Busby et al., 1993; Wilson et al., 1997).
A further aspect of the processing is to address the effects of the highly conductive fluid in the ST (the perilymph) and the relatively distant placements of the intracochlear electrodes from their neural targets (generally thought to be the spiral ganglion cells in the cochlea). The high conductivity and the
Spring 2019 | Acoustics Today | 55