| model: | |
| task: audio | |
| encoder: | |
| name: wavlm | |
| version: base | |
| pretrained: true | |
| pretrained_path: manipulate_model/encoder_checkpoints/wavlm/WavLM-Base+.pt | |
| output_layer: 3 | |
| encoder_freeze: false | |
| decoder: | |
| name: aasist | |
| version: default | |
| output_size: 2 | |
| online_encoding: true | |
| data: | |
| name: av1m | |
| train_parts: all | |
| val_parts: all | |
| test_parts: all | |
| train_size: -1 | |
| val_size: -1 | |
| test_size: -1 | |
| shape: | |
| - 3 | |
| - 224 | |
| - 224 | |
| sr: 16000 | |
| fps: 25 | |
| center_transition: true | |
| window_size: 4 | |
| sliding_window: false | |
| train: | |
| num_workers: 16 | |
| batch_size: 64 | |
| num_epochs: 15 | |
| optimizer: adam | |
| scheduler: step | |
| lr: 0.0001 | |
| step_size: 1 | |
| gamma: 0.1 | |
| loss: bce | |
| log_interval: 100 | |
| shuffle: true | |
| debug: false | |