This repository contains code which reproduces experiments presented in the paper Quaternion Convolutional Neural Networks for End-to-End Automatic Speech Recognition

Install requirements for experiments with pip:

pip install numpy tensorflow-gpu keras theano

Depending on your Python installation you might want to use anaconda or other tools. You can also go to the 'Installation' Section and execute the command for an automatic installation.

Install all the needed dependencies.

python setup.py install

Since the TIMIT dataset is not free, you must create your own features, and also create your own:

• Keras Data Generator (See scripts/training.py)
• Keras Edit Distance Accuracy (See scripts/training.py)

1. Get help:

   python scripts/run.py train --help
   

2. Run models:

   python scripts/run.py train --model {real,quaternion} --sf STARTFILTER --nl NUMBEROFLAYERS
   

   Other arguments may be added as well; Refer to run.py train --help for

   • Optimizer settings
   • Dropout rate
   • Clipping
   • Saving prefix
   • ...

Please cite our work as