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: 0.5313
• Wer: 0.3317

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
3.5823	1.0	500	1.8501	1.0236
0.8931	2.01	1000	0.5018	0.5196
0.4269	3.01	1500	0.4266	0.4461
0.2876	4.02	2000	0.4458	0.4359
0.2272	5.02	2500	0.4183	0.4146
0.1813	6.02	3000	0.4151	0.3945
0.1555	7.03	3500	0.4216	0.3881
0.1353	8.03	4000	0.4282	0.3824
0.1221	9.04	4500	0.4848	0.3845
0.1135	10.04	5000	0.5003	0.3818
0.0968	11.04	5500	0.5331	0.3738
0.09	12.05	6000	0.5082	0.3690
0.084	13.05	6500	0.4573	0.3634
0.0744	14.06	7000	0.4711	0.3705
0.0663	15.06	7500	0.4955	0.3634
0.0612	16.06	8000	0.4721	0.3558
0.0535	17.07	8500	0.4965	0.3654
0.0527	18.07	9000	0.5381	0.3592
0.0458	19.08	9500	0.5029	0.3498
0.0424	20.08	10000	0.5814	0.3547
0.042	21.08	10500	0.4893	0.3480
0.0373	22.09	11000	0.5047	0.3482
0.0333	23.09	11500	0.5235	0.3426
0.0306	24.1	12000	0.5165	0.3472
0.0293	25.1	12500	0.4988	0.3426
0.025	26.1	13000	0.5157	0.3382
0.0255	27.11	13500	0.5278	0.3412
0.022	28.11	14000	0.5401	0.3364
0.0195	29.12	14500	0.5313	0.3317

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 1.18.3
• Tokenizers 0.15.1