Training language models

MFA has a utility function for training ARPA-format ngram language models, as well as merging with a pre-existing model.

Steps to train:

  1. Provided the steps in Installation have been completed and you are in the same Conda/virtual environment that MFA was installed in.
  2. Run the following command, substituting the arguments with your own paths:
mfa train_lm corpus_directory output_model_path

Options available:


Display help message for the command

--config_path PATH

Path to a YAML config file for training the language model. see Language model configuration for more details.

--model_path PATH

Path to an existing language model to merge with the training data.

--model_weight WEIGHT

Specify the weight of the supplemental model when merging with the model from the training data.

--temp_directory DIRECTORY

Temporary directory root to use for aligning, default is ~/Documents/MFA