This is the source code for DEPA.

The code is centered around the config files placed at pretrain/config/config_*.yaml.

Use pretrain/data/raw_features.py scripts to extract LMS, STFT, or MFCC features for pretraining. You can modify pretrain/config/config_features.yaml to modify the configuration of features

The main script of this repo is pretrain/pretrain.py.

pretrain.py the following options ( ran as python pretrain.py OPTION):

• train: Trains a model given a config file (default is pretrain/config/config_pretrain.yaml).
• encoding: Extracting DEPA given model path and input file and output to the given output file.
• Pipeline: Conduct the above two steps

Example:

python pretrain.py train pretrain/config/config_pretrain.yaml False # mean not debug