Code for the paper "Contrastive Lyrics Alignment with a Timestamp-Informed Loss"
To install the required python packages, run
pip install -r requirements.txt
Refer to phonemizer on how to install the required background for IPA phonemes extraction.
The DALI dataset is used for training.
The jamendolyrics evaluation dataset is used for evaluation.
When you train for the first time, pickle files containing the training samples will be generated.
Run python train.py to train the model. Set the configurations in the config.py file.