danieleV9H/hubert-base-timit-demo-google-colab-ft30ep_v4

hubert-base-timit-demo-google-colab-ft35ep

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: 0.4602
• Wer: 0.3466

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
3.825	0.87	500	2.9521	1.0
2.431	1.73	1000	0.9760	0.8013
1.0089	2.6	1500	0.5934	0.5968
0.6859	3.46	2000	0.5132	0.5356
0.5302	4.33	2500	0.4506	0.4894
0.44	5.19	3000	0.4340	0.4670
0.3926	6.06	3500	0.4506	0.4528
0.3326	6.92	4000	0.4197	0.4486
0.2937	7.79	4500	0.4093	0.4193
0.2568	8.65	5000	0.4098	0.4229
0.2473	9.52	5500	0.4090	0.4141
0.2233	10.38	6000	0.4152	0.4125
0.2108	11.25	6500	0.4586	0.4189
0.2086	12.11	7000	0.4284	0.3969
0.1858	12.98	7500	0.4028	0.3946
0.1641	13.84	8000	0.4679	0.4002
0.1686	14.71	8500	0.4441	0.3936
0.1489	15.57	9000	0.4897	0.3828
0.1541	16.44	9500	0.4953	0.3783
0.1417	17.3	10000	0.4500	0.3758
0.1428	18.17	10500	0.4533	0.3796
0.1306	19.03	11000	0.4474	0.3792
0.1185	19.9	11500	0.4762	0.3743
0.1081	20.76	12000	0.4770	0.3699
0.1253	21.63	12500	0.4749	0.3629
0.1087	22.49	13000	0.4577	0.3534
0.1172	23.36	13500	0.4819	0.3525
0.1086	24.22	14000	0.4709	0.3623
0.089	25.09	14500	0.4852	0.3544
0.086	25.95	15000	0.4602	0.3555
0.086	26.82	15500	0.4861	0.3497
0.086	27.68	16000	0.4527	0.3473
0.0919	28.55	16500	0.4607	0.3487
0.0792	29.41	17000	0.4602	0.3466

Framework versions

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