How to train: STELA (audio) LSTM (phones or words) For BabyBERTa, please refer Philip Huebner's git repository.