Kaldi-Onboarding

Kaldi YesNo Tutorial GMM Training, Alignment and training a NNET 2 model

These scripts are largely based on points 2 and 3 under Some References.

It is advised that the reader or user goes line by line to get a proper understanding of what is happening behind the scenes

1. Execute ./run.sh This file will download the datasets and perform a training over monophones based on the downloaded dataset

2. Aligning Monophones, Generating and Aligning Triphones (Towards training a GMM model) Execute ./train_gmm.sh

3. Training a NNET-2 model For this GMM training must be done first, and there must be a aligned triphone directory present in the experiments directory Move the following files to "steps/nnet2"

   1. get_egs_simple.sh
   2. get_lda_simple.sh
   3. train_simple.sh
   4. decode_simple.sh Then Run ./run_nnet2_simple.sh

Some References:

1. GMM Training Overview: What essentially is happening when we are creating and aligning Monophones and Triphones: https://www.eleanorchodroff.com/tutorial/kaldi/training-overview.html

2. Steps involved in Training an Acoustic Model: https://www.eleanorchodroff.com/tutorial/kaldi/training-acoustic-models.html

3. JRMeyer's Tutorial on "How to Build a NN model in Kaldi": http://jrmeyer.github.io/asr/2016/12/15/DNN-AM-Kaldi.html

4. Kaldi-ASR Documentation (Onboarding Example): https://kaldi-asr.org/doc/tutorial.html

5. Kaldi-ASR Deep Neural Network Documentation: https://kaldi-asr.org/doc/dnn.html