The scripts in this repository can be used as a template for training nnet3 neural networks in Kaldi, with the aim to get you going from your data to a trained model as smoothly as possible.
The code here aims to be easily readable and extensible, and makes few assumptions about the kind of data you have and where it's located on disk.
To get started, easy-kaldi should be cloned and moved into the egs dir of your local version of the .
If you're used to typical Kaldi egs, take note that all easy-kaldi scripts in utils / local / steps exist in this repo. That is, they do not link back to the wsj example. This was done to make custom changes to the scripts, making them more readable.