An enhanced Conv-TasNet model for speech separation using a speaker distance-based loss function

J.A. Arango-Sánchez, and J.D. Arias-Londoño

Datasets

1. CallFriend-Caribbean-Spanish-Split
2. CallFriend-Spanish-Split
3. CallHome-Spanish-Corpus-Split

Train

1. python setup.py
2. python train_model.py --experiment_name "<>" --tags "<>" --save_best_model "<>" --default_root_dir "<>" --epochs <> --num_workers <> --root_csv "data/csv/local/" --batch_size <> --weight_CS <> --num_layers <>

Test

1. Download checkpoints
2. python test_model.py --experiment_name "<>" --tags "<>" --root_csv "<>" --save_best_model "MODEL_PATH"