THE AUDITORY MODELING TOOLBOX

Applies to version: 1.1.0

View the code

Go to function

KELVASA2015 - Azimuthal localization in cochlear-implant listeners

Usage

[results] = kelvasa2015(insig,fs,varargin);

Input parameters

insig Can be either [N x 2] two channel audio signal or results structure of preProcessed data
fs sampling rate (Hz)
varargin structure with all parameters required for model. If this is not included, default paramters are loaded.

Output parameters

results A structure containing the processed electrodograms, AN spike times, and model predicted azimuthal locations

Description

KELVASA2015(insig,fs) implements the ACE signal processing strategy upon the two channel input signal to produce bilateral electrodograms. This is further processed through an electrode nerve interface to generate spike times of a population of AN neurons. A chosen localization model from Kelvasa and Dietz 2015 is then used to map the two channel (right and left) outputs to a predicted azimuthal position.

WARNING: If Octave is used, the plots will differ from the original ones, because spectrogram is not available in Octave. Instead, sgram from LTFAT is used. The steps of the binaural model to calculate the result are the following :

1) Process two channel input signal through a CI strategy as detailed in (Hamacher, 2003) and (Fredelake and Hohmann, 2012) to produce bilateral electrodograms.

2) Process electrodogram through an electrode nerve interface and auditory nerve model as detailed in (Fredelake and Hohmann, 2012)

3) Compute bilateral spike rate differences over chosen AN frequency bands and time windows as detailed in (Kelvasa and Dietz, 2015)

4) Calibrate the chosen localization model with a chosen calibration signal. This step can take several hours so preProcessed calibration is loaded for "Speech Shaped Noise" at 55dB as detailed in (Kelvasa and Dietz, 2015)

5) Map the spike rate differences for each AN frequency band to a predicted azimuthal angle using the chosen localization model as detailed in (Kelvasa and Dietz, 2015)

The output structure "results" has the following fields:

electrodogramCHAN1
[NxM] matrix of CI electrode current output in mA(???) with N = number of CI electrodes and M = time
APvecCHAN1
[Nx2] matrix of [Nx1] indices of spiking AN fibers [Nx2] spike times in seconds
electrodogramCHAN2
same but for second channel
APvecCHAN2
same but for second channel
SpkSumPerBin
[NxM] matrix of Right and Left Spike Rate differences in spikes per second with N = number of AN frequency bands and M = time bins
SpkDiffPerBin
same but for Right and Left spike rate differences
ANbinPredictions
[NxM] matrix of azimuthal angle bin predictions in degrees with N= number of AN frequency bands and M = time bins
weightedPredictions
[1xM] matrix of bin weighted azimuthal angle bin predictions in degrees with M = time bins
mappingData
Structure containing data used to calibrate and implement the chosen localization model as detailed in kelvasa2015calibratemodels.m

References:

S. Fredelake and V. Hohmann. Factors affecting predicted speech intelligibility with cochlear implants in an auditory model for electrical stimulation. Hearing Research, 287(1):76 -- 90, 2012. [ DOI | http ]

V. Hamacher. Signalverarbeitungsmodelle des elektrisch stimulierten Gehörs; 1. Aufl. PhD thesis, RWTH Aachen, Aachen, 2004. Zugl.: Aachen, Techn. Hochsch., Diss., 2003. [ http ]

D. Kelvasa and M. Dietz. Auditory model-based sound direction estimation with bilateral cochlear implants. Trends in Hearing, 19:2331216515616378, 2015. [ DOI ]