train

Script to train causal and noncausal ConvTasNet on MUSDB18. Model is trained to separate the vocals from the background music.

recipes.cad2.task1.ConvTasNet.train.create_callbacks(conf: dict, exp_dir: str) → tuple

Create callbacks for the training.

recipes.cad2.task1.ConvTasNet.train.create_datasets_and_loaders(conf: dict) → tuple

Create train and val datasets and loaders.

recipes.cad2.task1.ConvTasNet.train.create_model_and_optimizer(conf: dict) → tuple

Create model, optimizer and lr_scheduler.

recipes.cad2.task1.ConvTasNet.train.create_trainer(conf: dict, callbacks: list) → Trainer

Create PyTorch Lightning trainer.

recipes.cad2.task1.ConvTasNet.train.get_loss_func() → L1Loss

Return the loss function.

recipes.cad2.task1.ConvTasNet.train.load_config(config_path)

recipes.cad2.task1.ConvTasNet.train.main(conf)

recipes.cad2.task1.ConvTasNet.train.save_best_model(system, checkpoint, exp_dir, train_set)