Skip on-the-fly Data Contamination steps


I am trying to use the templates provided in Google Colab to train an Acoustic Model for my own data. My data is already very noisy and I would not prefer to add any more env/background noise.

I think the best way to achieve that is to:

  1. change the values of all the parameters in “env_corrupt” to “0” in the hyperparameter-yaml file?


  1. comment the appropriate steps in prepare_features() & prepare_tokens() functions in Brain (ASR) class?

Can anyone verify whether I am correct? Appreciate your help! :smiley:

Hi ! Simply comment it on the yaml and it will be deactivated :slight_smile:

1 Like

Thanks, appreciate your help! :+1:

1 Like