Binaural Unmasking with Bilateral Cochlear Implants

Abstract

Nearly 100,000 deaf patients worldwide have had their hearing restored by a cochlear implant (CI) fitted to one ear. However, although many patients understand speech well in quiet, even the most successful experience difficulty in noisy situations. In contrast, normalhearing (NH) listeners achieve improved speech understanding in noise by processing the differences between the waveforms reaching the two ears. Here we show that a form of binaural processing can be achieved by patients fitted with an implant in each ear, leading to substantial improvements in signal detection in the presence of competing sounds. The stimulus in each ear consisted of a narrowband noise masker, to which a tonal signal was sometimes added; this mixture was half-wave rectified, lowpass-filtered, and then used to modulate a 1000-pps biphasic pulse train. All four CI users tested showed significantly better signal detection when the signal was presented out of phase at the two ears than when it was in phase. This advantage occurred even though subjects only received information about the slowly varying sound envelope to be presented, contrary to previous reports that waveform fine structure dominates binaural processing. If this advantage generalizes to multichannel situations, it would demonstrate that envelope-based CI speech-processing strategies may allow patients to exploit binaural unmasking in order to improve speech understanding in noise. Furthermore, because the tested patients had been deprived of binaural hearing for eight or more years, our results show that some sensitivity to time-varying interaural cues can persist over extended periods of binaural deprivation.

Sensitivity to interaural envelope correlation changes in bilateral cochlear-implant users

“…Omitting the one listener who did not have a determinable ITD JND, a linear regression between the 50-Hz NoSp thresholds and ITD JNDs accounted for 67% (p ¼ 0.007) of the variance in the data. A similar significant relationship between NoSp thresholds and ITD JNDs was found by Long et al (2007). Third, we measured psychometric functions, whereas Lu et al performed an adaptive tracking procedure to determine threshold.…”

“…The average BMLD for tones in noise measured here were found to be larger than the average BMLDs for tones in noise measured by Lu et al (2010) and were more similar to the BMLDs for transposed tones measured by Long et al (2006) and Van Deun et al (2011). Although our stimuli are most like those in Lu et al (there were no regular gaps in stimulation as occur for transposed tones), several factors might explain the discrepancy between our data and those of Lu et al First, we provided feedback and training prior to collecting data, whereas Lu et al did not provide training.…”

“…The envelope was then sampled at equal intervals corresponding to 1000 pps. The envelopes were compressed by a function similar to the one used by Long et al (2006), which was an approximation of the compression used by Cochleartype speech processors. This compression function is designed to transform the typically highly expansive loudness growth function of a CI listener to a loudness growth function that is more similar to a NH listener.…”

“…To summarize, the analog amplitudes were compressed, quantized, and placed on a CU scale between T and C, and levels 30 dB below the average peak envelope amplitude were set to zero. Note that unlike Long et al (2006) and Lu et al (2010), the analog stimuli were not normalized to produce a peak amplitude that was unity for every token. Instead the analog stimuli were normalized such that the average peak amplitude over the 25 tokens corresponded to unity amplitude.…”

“…Several studies with synchronized processors have also shown that bilateral CI users demonstrate positive BMLDs. That is, thresholds are lower for conditions with NoSp (diotic noise, tone out-of-phase) or NoSs (diotic noise, tone with a nonzero ITD) compared to NoSo (diotic noise, tone in-phase and zero ITD) (Long et al, 2006;Van Deun et al, 2009;Lu et al, 2010;Lu et al, 2011;Van Deun et al, 2011). Long et al measured average BMLDs of 29 dB for 125-Hz transposed tones (i.e., stimulation pulses were dropped at regular intervals to produce small gaps in stimulation) using single pitch-matched pairs of electrodes in four CI listeners.…”