ThunderHopper

Model of the auditory pathway of grasshoppers.

Auditory pathway

Acoustic signals are sequentially processed along the auditory pathway:

1. Raw auditory signal (raw).
2. Bandpass filterd by tympanum (filt).
3. Computation of envelope by rectification and low-pass filtering (env).
4. Logarithmic transform into decibel (log).
5. High-pass filtering to generate intensity invariant envelope (inv).
6. Convolution of the envelope with a set of Gabor kerneles (conv).
7. Thresholding the convolved traces (bi).
8. Low-pass filtering the binary trace to generate slowly varying features (feat).

In brackets the acronyms are given that are used by output dictionaries and for signal selection as described below.

Usage

First, import what we need:

import numpy as np
import matplotlib.pyplot as plt
from thunderhopper import configuration, process_signal, load_data

Second, prepare a configuration dictionary. You simply need to provide the standard deviations of the Garbor kernels and their types. The latter is the number of lobes you want, a negative number flips the kernel on the x-axis:

sigmas = [0.001, 0.002, 0.004, 0.008, 0.016, 0.032]
types = [1, -1, 2, -2, 3, -3, 4, -4, 5, -5,
         6, -6, 7, -7, 8, -8, 9, -9, 10, -10]
config = configuration(types=types, sigmas=sigmas)

Then just run the model on an audio recording:

data, rates = process_signal(config, path='recording.wav')

That's it!

Both, data and rates are dictionaries with the same keys. data contains the time series and rates the corresponding sampling rates. To plot the filtered signal and the envelope of the first channel just run

filt = data['filt'][:, 0]
filt_rate = rates['filt']
tfilt = np.arange(len(filt))/filt_rate

env = data['env'][:, 0]
env_rate = rates['env']
tenv = np.arange(len(env))/env_rate

plt.plot(tfilt, filt)
plt.plot(tenv, env)
plt.show()

Save model traces in file

You can also save all computed traces in a numpy npz file. Just pass a file path to the save argument:

process_signal(config, path=data_path, save='recording.npz')

Now, all the computed traces are saved in this file.

Load this file like this:

data, params = load_data('recording.npz', ['filt', 'env'])
filt_rate = params['rate']
env_rate = params['env_rate']

The second argument requests only the filtered signal and the envelope to be loaded from the file. When not specified, all traces are loaded.

Configuration parameters

After calling configuration(), you may change some of the parameters of the model.

For example, you can set individual thresholds for each feature by loading them from a file and change the low-pass filter's cutoff frequency that generates the features:

config.update({
    'feat_thresh': np.load('acrididae.npy') * 0.1,
    'feat_fcut': 0.75,
    })

Or run the song detection on a specific input channel with a specific threshold:

config.update({
    'label_channels': 0,
    'label_thresh': 0.5
    })

Select what to store

If you do not need the traces from every step of the auditory pathway, you may select what to store via the returns argument. For example, if you are only interested in the features, the norm of the features and the song labels, call process_signal() like this:

returns = ['feat', 'norm', 'songs']
data, rates = process_signal(config, path='recording.wav', returns=returns)