Team Lightning's solution for Zalo AI Challenge 2022
Link:
https://drive.google.com/file/d/1vUSxdORDxk-ePUt-bUVDahpoXiqKchMx/
Note that users should create a metadata file in .json format with audio_filepath, transcript and note metadata fields.
pip install -r requirements.txt
We pre-trained the model with self-supervised approach on Vin BigData ASR
python train.py --config-name pretrain.yaml
After that, this command is to fine-tune the model with ZAC2022 train dataset (Optional)
python train.py --config-name finetune.yaml
Model Inference can be simply performed with one line
python inference.py