A python console application that enables machine learning on directional room impulse responses from microphone array recordings or simulations.

It is recommended to create and activate a new and empty conda environment:

conda create --name myenv python=3.6
source activate myenv # linux, mac
activate myenv # windows

Clone the project:

git clone https://github.com/Agent49/drirlearning.git

Install the requirements:

pip install -r requirements.txt

Tensorflow requires python 3.4 to 3.6. It is highly recommended to run the application with GPU-support. For more information see:

https://www.tensorflow.org/install/gpu

You can run the application from console. Optional parameters are useful if you want to quickly switch from small to huge data sets or change some hyperparameters. For more information type:

python ./drirlearning.py -h

The best way to assess and compare the performance of your models is a visual inspection of loss and other results with TensorBoard. Therefore, the last line on your console gives you instructions after the process has finished.

The purpose of this application is that you create your own models/neural nets and train them on different data sets. Initial models and sample data is provided. The principle steps are as follows:

1. Create your own model in model.py.
2. In drirlearning.run() call function utils.run_model(your_model...), given your models as a callback function.
3. Adjust the configuration hard-coded or via CLI interface.
4. Run the application.

The modul drirlearning.utils.py will provide you with helper functions. You can build the complete documentation for your browser with search functions by typing:

cd ./docs
make html