hubert-base-timit-demo-google-colab-ft30ep_v5

This model is a fine-tuned version of facebook/hubert-base-ls960 on the timit-asr dataset. It achieves the following results on the evaluation set:

• Loss: 0.4763
• Wer: 0.3322

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.9596	0.87	500	3.1237	1.0
2.5388	1.73	1000	1.1689	0.9184
1.0448	2.6	1500	0.6106	0.5878
0.6793	3.46	2000	0.4912	0.5200
0.5234	4.33	2500	0.4529	0.4798
0.4368	5.19	3000	0.4239	0.4543
0.3839	6.06	3500	0.4326	0.4339
0.3315	6.92	4000	0.4265	0.4173
0.2878	7.79	4500	0.4304	0.4068
0.25	8.65	5000	0.4130	0.3940
0.242	9.52	5500	0.4310	0.3938
0.2182	10.38	6000	0.4204	0.3843
0.2063	11.25	6500	0.4449	0.3816
0.2099	12.11	7000	0.4016	0.3681
0.1795	12.98	7500	0.4027	0.3647
0.1604	13.84	8000	0.4294	0.3664
0.1683	14.71	8500	0.4412	0.3661
0.1452	15.57	9000	0.4484	0.3588
0.1491	16.44	9500	0.4508	0.3515
0.1388	17.3	10000	0.4240	0.3518
0.1399	18.17	10500	0.4605	0.3513
0.1265	19.03	11000	0.4412	0.3485
0.1137	19.9	11500	0.4520	0.3467
0.106	20.76	12000	0.4873	0.3426
0.1243	21.63	12500	0.4456	0.3396
0.1055	22.49	13000	0.4819	0.3406
0.1124	23.36	13500	0.4613	0.3391
0.1064	24.22	14000	0.4842	0.3430
0.0875	25.09	14500	0.4661	0.3348
0.086	25.95	15000	0.4724	0.3371
0.0842	26.82	15500	0.4982	0.3381
0.0834	27.68	16000	0.4856	0.3337
0.0918	28.55	16500	0.4783	0.3344
0.0773	29.41	17000	0.4763	0.3322

Framework versions

• Transformers 4.17.0
• Pytorch 1.11.0+cu113
• Datasets 1.18.3
• Tokenizers 0.12.1