| { | |
| "name": "test_config", | |
| "n_gpu": 1, | |
| "preprocessing": { | |
| "sr": 16000, | |
| "spectrogram": { | |
| "type": "MelSpectrogram", | |
| "args": { | |
| } | |
| }, | |
| "log_spec": true | |
| }, | |
| "augmentations": { | |
| "wave": [], | |
| "spectrogram": [] | |
| }, | |
| "arch": { | |
| "type": "BaselineModel", | |
| "args": { | |
| "n_feats": 128, | |
| "fc_hidden": 512 | |
| } | |
| }, | |
| "data": { | |
| "train": { | |
| "batch_size": 20, | |
| "num_workers": 0, | |
| "datasets": [ | |
| { | |
| "type": "LibrispeechDataset", | |
| "args": { | |
| "part": "dev-clean", | |
| "max_audio_length": 20.0, | |
| "max_text_length": 200 | |
| } | |
| } | |
| ] | |
| }, | |
| "val": { | |
| "batch_size": 20, | |
| "num_workers": 0, | |
| "datasets": [ | |
| { | |
| "type": "LibrispeechDataset", | |
| "args": { | |
| "part": "dev-clean", | |
| "max_audio_length": 20.0, | |
| "max_text_length": 200 | |
| } | |
| } | |
| ] | |
| } | |
| }, | |
| "optimizer": { | |
| "type": "SGD", | |
| "args": { | |
| "lr": 3e-4 | |
| } | |
| }, | |
| "loss": { | |
| "type": "CTCLoss", | |
| "args": {} | |
| }, | |
| "metrics": [ | |
| { | |
| "type": "ArgmaxWERMetric", | |
| "args": { | |
| "name": "WER (argmax)" | |
| } | |
| }, | |
| { | |
| "type": "ArgmaxCERMetric", | |
| "args": { | |
| "name": "CER (argmax)" | |
| } | |
| } | |
| ], | |
| "lr_scheduler": { | |
| "type": "OneCycleLR", | |
| "args": { | |
| "steps_per_epoch": 100, | |
| "epochs": 50, | |
| "anneal_strategy": "cos", | |
| "max_lr": 4e-3, | |
| "pct_start": 0.2 | |
| } | |
| }, | |
| "trainer": { | |
| "epochs": 50, | |
| "save_dir": "saved/", | |
| "save_period": 5, | |
| "verbosity": 2, | |
| "monitor": "min val_loss", | |
| "early_stop": 100, | |
| "visualize": "wandb", | |
| "wandb_project": "asr_project", | |
| "len_epoch": 100, | |
| "grad_norm_clip": 10 | |
| } | |
| } | |