Practiced implementing a simple DNN (along with discriminative pretraining) for acoustic modelling for ASR and used kaldi for frontend and decode. Kaldi features, alignment, etc are used in python using Karel's (https://github.com/vesis84) library (https://github.com/vesis84/kaldi-io-for-python).
Add documentation and comments