inverse_cochlea

inverse_cochlea can reconstruct sounds from the activity of auditory nerve fibers' using artificial neural networks:

__|______|______|____    +-----------+
_|________|______|___ -->|  Inverse  |      .-.     .-.     .-.
___|______|____|_____ -->|           |-->  /   \   /   \   /   \
__|______|______|____ -->|  Cochlea  |          '-'     '-'
                         +-----------+
      ANF activity                                 Sound

Bernstein Conference 2012

Rudnicki M, Zuffo MK and Hemmert W (2012). Sound Decoding from Auditory Nerve Activity. Front. Comput. Neurosci. Conference Abstract: Bernstein Conference 2012. doi: 10.3389/conf.fncom.2012.55.00092

Full abstract

Sound samples

Poster

Usage

The direct reconstruction with an artificial neural network (suitable for frequencies to up 2 kHz) is implemented in inverse_cochlea.MlpReconstructor. The reconstruction using a combination of an artificial neural network and inverse spectrogram is implemented in inverse_cochlea.ISgramReconstructor.

Both reconstructor classes can be configured by the constructor parameters and have train() and run() methods.

To see how to use the package, have a look at the scripts in the examples directory.

Requirements

• Python 2.7
• Numpy
• Scipy
• Pandas
• joblib
• cochlea

For the MlpReconstructor:

• ffnet

For the ISgramReconstructor:

• oct2py or pytave
• GNU Octave with ltfat preinstalled

License

The project is licensed under the GNU General Public License v3 or later (GPLv3+).