Verhulstetal2018Model

The model code version 1.2 for the Verhulst, Altoè, Vasilkov 2018 Hearing Research publication: Computational modeling of the human auditory periphery: Auditory-nerve responses, evoked potentials and hearing loss. Hearing Research 360, 55-75. (found in 'doc/' folder)

A short and an extended version of the v1.2 release note can be found in the 'doc/' folder. The previous version can be found under the v1.1-master git branch. Version 1.2 includes an update of the Inferior Colliculus stage (IC). A minor change was introduced to the cochlear nucleus stage (CN). As a consequence, the scaling constants M1, M3, and M5 (calibrated using the flag 'abr') were also updated. All other model stages remained unchanged.

This work received funding from the European Research Council (ERC) under the Horizon 2020 Research and Innovation Programme (grant agreement No 678120 RobSpear)

The model code and interface was written by Alessandro Altoè and Sarah Verhulst (copyright 2012,2014,2015,2016,2018) and is licensed under the UGent acadamic license (see details in license file that is part of this repository). The Verhulstetal2018Model consists of the following files: tridiag.so, cochlea_utils.c, build.bat, build.sh, run_model2018.py, model2018.m, cochlear_model2018.py, inner_hair_cell2018.py, auditory_nerve2018.py, ic_cn2018.py, ExampleSimulation.m, ExampleAnalysis.m, the HI profiles in the Poles folder. Changes in the IC/CN stages (version 1.2) were introduced by Alejandro Osses Vecchi and Sarah Verhulst.

##################### How to run the model #####################

INSTALL NUMPY AND SCIPY (anaconda), check whether you should install 32 or 64 bit! The model works on python 2.7 and also on 3.6 (with some future warnings) verify with some simple code or the "pyversion" command whether python can be called from the MATLAB command line (check the internet for examples)

COMPILE THE tridiag.so or tridiag.dll (in case you use Windows) file, i.e. the tridiagonal matrix solver part of the cochlear mechanics

2.1 for mac/linux:

 open a terminal, go to the model folder and type:
     gcc -shared -fpic -O3 -ffast-math -o tridiag.so cochlea_utils.c
 OR run build.sh script:
     ./build.sh

2.2 for windows:

 The native GCC compiler from the MinGW project needs to be installed on a Windows machine. It can be downloaded from here: https://sourceforge.net/projects/mingw-w64/ (other choices such as cygwin can also work)
 In the installation window choose the desired compiler version (e.g. the latest one) and the computer architecture (i686 for 32-bit OS or x86_64 for 64-bit OS). Install the compiler.
 After the installation is complete, run the migw-w64.bat file from the installation directory (e.g. C:\Program Files\mingw-w64\x86_64-8.1.0-win32-seh-rt_v6-rev0\).
 Via the terminal go to the model’s folder.
 Run the command:
     gcc --version (to check if gcc was successfully installed)
     gcc -shared -fpic -O3 -ffast-math -o tridiag.dll cochlea_utils.c
 OR run build.bat script

2.3 For Windows using Windows Subsystem for Linux

 If you already use Windows Subsystem for Linux and have Ubuntu installed, the following will generate a 64-bit dll::

     sudo apt install mingw-w64
     x86_64-w64-mingw32-gcc -shared -fpic -O3 --fast-math -o tridiag.dll cochlea_utils.c

 For a 32-bit dll, use i686-w64-mingw32-gcc instead.

Unzip the "Poles" folder
RUN THE MATLAB SCRIPT: ExampleSimulation.m When you run the model and everything works: "running human auditory model 2018 (version 1.2): Verhulst, Altoe, Vasilkov" printed is in the command window. Depending on the number of CPU cores you have, you can usually run 6-10 stimuli at once (i.e., size of the stim vector) Keep the stimuli short! 100-200ms stimuli work well. The model does not crash for long stimuli, but it may take very long before it is ready. If you want to run speech, please start with shorter segments first and find your optimal approach. Note that you may have to limit the number of parameters you save to optimize disk space, if you run long simulations.
RUN THE MATLAB SCRIPT: ExampleAnalysis.m
NOTE TO PYTHON FANS: Because all the model code is written in python, it is possible to run the model without MATLAB. MATLAB is only used here to interface with the model: design the stimuli, set the parameters and plot the results.
MAKING MODIFICATIONS BEYOND THE STANDARD PARAMETERS

7.1 The "Poles" folder has a whole range of auditory profiles (cochlear gain loss) that can be simulated. Each folder corresponds to a specific audiogram shape: FlatXX refers to a fixed dB HL loss across CF. SlopeXX refers to a sloping HF loss starting from 1 kHz and XX corresponds to the loss in dB HL at 8 kHz. SlopeXX_Y refers to a sloping HF loss from 1 kHz and a fixed Y dB HL loss for CFs below 1 kHz. In each folder, you find the alpha*,A values that should be loaded into the model (i.e. the StartingPoles.dat file for the considered HL profile). The audiogram shape can be seen by plotting the first and second line of the profile.txt file against eachother. The Poles.mat file has the HI starting poles and corresponding QERBs across the frequencies in fres as well as the the NH reference poles and QERBs.

7.2 The middle-ear filter parameters can be changed in line 226-227 of cochlear_model2018.py

7.3 The cochlear compression slope can be changed in line 414 of cochlear_model2018.py by changing the 0.31 to another value self.PoleE = np.zeros_like(self.x)+0.31 #saturating pole note that the "compression_slope=0.4" parameter in line 145 is NOT active (it came from earlier version)

7.4 The cochlear irregularity percentage (for reflection-source emissions) can be changed in lines 146 and 147 in cochlear_model2018.py by modifying the percentage=0.05 (reflection-source strenght) and kneevar=1 (horizontal random dB shift of vbm,30 compresssion kneepoint)

7.4 The stimulus level at which the nonlinearity kicks (i.e. the vbm at which compression starts, vbm,30) cannot be changed easily, the vbm thresholds need to be derived from simulations using linear models. It is currently set to a fixed BM vel/disp value corresponding to 30-dB pure-tone response at 1 kHz (see line 415,416).

7.5 Simulating reflection-source emissions should be done in two simulations: a) put the irregularities on (has reflection-source and distortion-source) b) put the irregularities off (has distortion-source) subtract the emission waveforms.

Happy Modeling!

Sarah and the Hearing Technology group

########### References ###########

Please cite the 2018 Hearing Research paper and other relevant papers when you use the model (or parts of it) for your research. Additional references with information on model specifics and earlier implementations can be found here:

Altoè A, Pulkki V, Verhulst S. (2018) The effects of the activation of the inner-hair-cell basolateral K+ channels on auditory nerve responses. Hearing Research 364: 64-80. => Shows how adding the K+ channels in IHC transduction model parameters, has important consequences for the adaptation properties of the auditory nerve.

Altoè A, Pulkki V, Verhulst S. (2017) Model-based estimation of the frequency-tuning of the inner hair cell stereocilia from neural tuning curves. Journal of the Acoustical Society of America, 141 => Explains the IHC transduction model parameters.

Verhulst S, Jagadeesh A, Mauermann M, Ernst F (2016). Individual differences in auditory brainstem response wave characteristics: Relations to different aspects of peripheral hearing loss. Trends in Hearing 20, 2331216516672186. => Application of the ABR model for hearing-impairment and synaptopathy (HI pole profiles)

Verhulst S, Bharadwaj H, Mehraei G, Shera CA, Shinn-Cunningham, BG. (2015). Functional modeling of the human auditory brainstem response to broadband stimulation. Journal of the Acoustical Society of America 138 (3): 1637-1659. => The CF dependent SheraPoles calibrated to human SFOAE tuning across frequency is described here => updated ME filter parameters. => note that it is better to use the 2018 model when simulating OAEs given the reactive cochlear Z0 in the 2015 implementation that is now resistive in the 2018 cochlea.

Altoe A, Pulkki V, Verhulst S. (2014). Transmission-line Cochlear Models: Improved Accuracy and Efficiency. Journal of the Acoustical Society of America, 136 EL302. => The equation solvers and stability is explained here

Verhulst S, Dau T, Shera CA. (2012). Nonlinear time-domain cochlear model for transient stimulation and human otoacoustic emission. Journal of the Acoustical Society of America, 132 (6), 3842-3848. => Explains the cochlear mechanics and moving pole nonlinearity concept

Saremi A, Beutelmann R, Dietz M, Ashida G, Kretzberg J and Verhulst S (2016). A comparative study of seven human cochlear filter models. Journal of the Acoustical Society of America, 140(3), pp.1618-1634. => Shows cochlear mechanical responses to different stimuli and different filterbank models

bburan/Verhulstetal2018Model

Verhulstetal2018Model