Cochlear implants (CI) are now in widespread use and the average performance level is excellent, but there is considerable variability in performance across patients. The 1995 NIH Consensus Conference on Cochlear Implants identified variability in patient outcomes as one of the key areas in need of further research. Recent research in our lab has shown little variability in performance across normal-hearing listeners tested with the same signal processing as CI listeners. Our overall hypothesis is that this variability is partly (or even largely) due to poor fitting of the processor parameters to the individual patient, particularly in the frequency domain. Experiments on normal-hearing listeners show that warping the frequency-to- cochlear place assignments can result in a large degradation in performance. Our overall goal is to develop a quantitative model of the effects of frequency-place mismatching, which could be used to guide the fitting of speech processors in cochlear implants and hearing aids.
Specific Aim 1 is to quantify the effects of various types of distortion in the frequency-place mapping on speech recognition in quiet and in noise. Distortions to be characterized are: frequency-place shifts, truncation in acoustic or cochlear domains, and linear and nonlinear frequency-place warping. We will measure phoneme, word and sentence cognition as a function of the signal-to-noise ratio [performance-intensity (PI) functions]. Simple sigmoidal functions will be fit to the PI functions and the three parameters [speech recognition threshold (SRT), slope, and asymptote] will be estimated as a function of the independent variable.
Specific Aim 2 is to assess the differences between CIS, SAS, and n-of-m processing strategies. One experiment will directly compare performance in the same NH and CI listeners for CIS and n-of-m processors. Another experiment will assess place-specific temporal processing in SAS processors and the results will be compared to CIS and n-of-m processing with the same frequency- place map.
Specific Aim 3 is to develop a mathematical model of the effects of various types of distortions in the frequency-place mapping, based on the data obtained in Aim 1. This model would define a multidimensional surface relating speech recognition as a function of signal-to-noise level to frequency-place mapping parameters. The model would provide a quantitative description of this parameter space and would provide the basis for frequency-place parameter optimization strategies. Parameter optimization procedures (Simplex, Neural Networks, Genetic algorithms) will then be applied to this surface to find the frequency-place (or frequency-electrode) assignment that produces maximum performance.
Showing the most recent 10 out of 43 publications
