Corrective binaural processing for bilateral cochlear implant patients

Abstract

Although bilateral cochlear implant users receive input to both ears, they nonetheless have relatively poor localization abilities in the horizontal plane. This is likely because of the two binaural cues, they have good sensitivity to interaural differences of level (inter-aural level differences, or ILDs), but not those of time (inter-aural time differences; ITDs). Here, localization performance is assessed in six bilateral cochlear implant patients when instantaneous ITDs are measured and converted to ILDs, a strategy that results in larger-than-typical ILDs. The added ILDs are corrective, in that they are derived from individual listener performance across both frequency and azimuth, so that they are small where a listener performs well, and increase as performance deviates from ideal. Results show significantly improved localization performance as a result of this strategy, with two of the six listeners achieving levels of performance typically observed in NH listeners.

Fig 1. Corrective ILD generation data for participant 883.

The six inset panels each represent a different frequency band, with cutoff frequencies indicated in the upper left corner of each inset panel. In each inset panel, the dashed black line represents ideal performance, the open blue circle markers depict perceived location plotted as a function of source location, and the heavy blue plot is a sigmoid fit to the localization data. The red functions are the corrective ILD functions for each band, which are the respective differences between ideal and the fit sigmoids, are expressed in dB, and use the y-axes to the right.

Fig 2. Corrective ILD generation data for participant 2243.

The layout is identical to that of Fig 1.

Fig 3. Corrective ILD generation data for participant 2287.

The layout is identical to that of Fig 1.

Fig 4. Corrective ILD generation data for participant 1363.

The layout is identical to that of Fig 1.

Fig 5. Corrective ILD generation data for participant 643.

The layout is identical to that of Fig 1.

Fig 6. Corrective ILD generation data for participant 2349.

The layout is identical to that of Fig 1.

Fig 7. Localization performance for each participant.

Each of the six participants are each represented in a separate panel. The blue open circles represent performance in the Unprocessed condition, the red open circles represent performance in the Processed condition, and error bars represent standard deviations. The X axes are staggered within each panel to facilitate comparison and to conserve space.

Fig 8. ITD frequency histograms for various frequencies.

Acoustic analysis was conducted on 50 pairs of sentences (about 2.5 minutes of data) produced by different female talkers, in which head-related impulse responses were applied to each such that one was 60 degrees to the left, and the other 60 degrees to the right. The sentences were then combined, filtered into six one-octave wide frequency bands, and an ITD was then estimated every 20 ms in each band. Each plot represents the proportion of occurrence of a given ITD among the time bins. Each panel shows acoustic analysis data for a particular frequency band.

Fig 9. Pre- and post-processed ILD estimates as a function of azimuth for each participant.

Acoustic analysis is of 500-ms broadband Gaussian noise, recorded from source locations at +/-90 degrees in 15-degree steps, using omnidirectional 1/4” microphones mounted on an acoustic manikin facing forward just above the auricles, as was done during testing. The top panel depicts ILDs in the unprocessed condition, and the the lower six panels each represent ILDs after processing using the six listeners corrective ILD functions, one panel per listener. In each panel, maker size represents ILD magnitude in each of the six frequency bands used in the study as a function of source location. Black markers represent apparent source locations to the left, and white markers to the right.