wav2vec2-base-timit-demo-google-colab

This model is a fine-tuned version of facebook/wav2vec2-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5155
• Wer: 0.3388

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
3.5822	1.0	500	2.4127	1.0
0.9838	2.01	1000	0.5401	0.5363
0.4308	3.01	1500	0.4380	0.4592
0.3086	4.02	2000	0.4409	0.4503
0.2324	5.02	2500	0.4148	0.4041
0.202	6.02	3000	0.4214	0.3882
0.1595	7.03	3500	0.4489	0.3875
0.1383	8.03	4000	0.4225	0.3858
0.1246	9.04	4500	0.4512	0.3846
0.104	10.04	5000	0.4676	0.3875
0.0949	11.04	5500	0.4389	0.3683
0.0899	12.05	6000	0.4964	0.3803
0.0854	13.05	6500	0.5397	0.3798
0.0728	14.06	7000	0.4823	0.3666
0.065	15.06	7500	0.5187	0.3648
0.0573	16.06	8000	0.5378	0.3715
0.0546	17.07	8500	0.5239	0.3705
0.0573	18.07	9000	0.5094	0.3554
0.0478	19.08	9500	0.5334	0.3657
0.0673	20.08	10000	0.5300	0.3528
0.0434	21.08	10500	0.5314	0.3528
0.0363	22.09	11000	0.5540	0.3512
0.0326	23.09	11500	0.5514	0.3510
0.0332	24.1	12000	0.5439	0.3492
0.0275	25.1	12500	0.5273	0.3432
0.0267	26.1	13000	0.5068	0.3430
0.0243	27.11	13500	0.5131	0.3388
0.0228	28.11	14000	0.5247	0.3406
0.0227	29.12	14500	0.5155	0.3388

Framework versions

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