Introduction

PyZBC2014 is a barebones Python 3 wrapper around the Zilany, Bruce, and Carney (2014) auditory-nerve (AN) model, which is written in C. Currently, PyZBC2014 is set up as a Python package (i.e., it can be imported like using pyzbc2014 as zbc in Python scripts), but installing PyZBC2014 currently requires some manual steps.

PyZBC2014 is licensed under the GNU AGPLv3 license (see LICENSE). Please send any questions or issues to daniel_guest@urmc.rochester.edu, or raise an issue on the GitHub page.

Install

Download

To install PyZBC2014, first download (and, if needed, unpack) the files from GitHub page.

Compile model

Next, compile the .c files that are bundled with the package by navigating to the pyzbc2014/model folder and running the following code for GCC (or equivalent code, based on your C compiler). If you are successful, you should see a file called libzbc2014.so in the model folder, alongside .c, .o, and some other files.

Windows (with gcc)

Installing and using C compilers on Windows can be a pain; you may want to try https://www.msys2.org/. With gcc installed, the following command will compile the model:

gcc -fPIC -O3 -shared -o libzbc2014.so complex.c model_IHC.c model_Synapse.c

Install package

Finally, to install the package, activate your desired Python environment and, in the top-level folder, use:

pip install .

Adjust pip to pip3 or others as needed based on your Python environment configuration. If installation succeeds, you should now be able to import pyzbc2014 as zbc

Uninstall package

Use pip to uninstall:

pip uninstall pyzbc2014

Usage

PyZBC2014 defines two function, sim_ihc_zbc2014 and sim_anrate_zbc2014, to simulate inner-hair-cell (IHC) potentials from sound-pressure waveforms or AN instantaneous rates from IHC potentials, respectively. Comparing to the origianl MATLAB/Mex implementation of the 2014 model, these behave similarly to the IHCAN and Synapse functions, respectively, with some small differences in behavior and some improvements in quality-of-life.

Simulate IHC response

To simulate an IHC response, you need to provide a sound-pressure waveform in 1D Numpy vector format to the sim_ihc_zbc2014 function; it will return an IHC potential waveform of exactly the same size and format. Parameters can be modified by keyword arguments:

• cf: Characteristic frequency (Hz)
• nrep: Number of reps to generate (only a single rep is simulated, others are appended copies)
• fs: Sampling rate (Hz), should generally be at least 100 kHz, and higher for cat/very-high-CF simulations; note that your stimulus must be sampled at this frequency!
• cohc: Status of the outer hair cells, in range of [0, 1] with 1 being fully normal/healthy
• cihc: Status of the inner hair cells, in range of [0, 1] with 1 being fully normal/healthy
• species: String encoding species to use in ["cat", "human", "human-glasberg"]. In the MATLAB/Mex wrapper, species=1 corresponds to "cat", species=2 corresponds to "human" (and indicates human with the sharp peripheral tuning suggested by Shera and Oxenham), and species=3 corresponds to "human-glasberg" (and indicates broader peripheral tuning suggested by Glasberg and Moore)

The full function call using default keyword arguments is thus:

ihc_potential = sim_ihc_zbc2014(px, cf=1e3, nrep=1, fs=100e3, cohc=1.0, cihc=1.0, species="human")

Simulate AN response

To simulate an AN response, you need to provide an IHC potential (from the output of sim_ihc_zbc2014) in 1D Numpy vector format to the sim_anrate_zbc2014 function; it will return a prediction of instantaneous AN firing rate rate of exactly the same size and format. Parameters can be modified by keyword arguments:

• cf: Characteristic frequency (Hz)
• nrep: Number of reps to simulate, should always match value used in sim_ihc_zbc2014
• fs: Sampling rate (Hz), should generally be at least 100 kHz, and higher for cat/very-high-CF simulations, should always match value used for sim_ihc_zbc2014
• fibertype: String encoding which fiber type to simulate in ["hsr", "msr", "lsr"]
• powerlaw: String encoding whether to use true powerlaw adaptation ("true") or approximate powerlaw adaptation ("approx"). In the MATLAB/Mex wrapper, implnt=1 corresponds to "true" and implnt=0 corresponds to "approx"
• noisetype: String encoding whether to use no fractional Gaussian noise ("none"), or fresh noise ("fresh"). In the MATLAB/Mex wrapper, noiseType=1 corresponds to "fresh"

The full function call using default keyword arguments is thus:

anrate = sim_anrate_zbc2014(ihc, cf=1e3, nrep=1, fs=100e3, fibertype="hsr", powerlaw="approx", noisetype="fresh")