Online decoding in Kaldi uses GMM (or NN+GMM hybrid) where this repo focus in end to end approach using deep LSTM (similar to Google Pixel's offline speech to text [1]).

[1] https://arxiv.org/pdf/1811.06621.pdf