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: 0.5436
• Wer: 0.3401

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
3.5276	1.0	500	1.9983	1.0066
0.8606	2.01	1000	0.5323	0.5220
0.4339	3.01	1500	0.4697	0.4512
0.3026	4.02	2000	0.4342	0.4266
0.2297	5.02	2500	0.5001	0.4135
0.1939	6.02	3000	0.4350	0.3897
0.1613	7.03	3500	0.4740	0.3883
0.1452	8.03	4000	0.4289	0.3825
0.1362	9.04	4500	0.4721	0.3927
0.1146	10.04	5000	0.4707	0.3730
0.1061	11.04	5500	0.4470	0.3701
0.0947	12.05	6000	0.4694	0.3722
0.0852	13.05	6500	0.5222	0.3733
0.0741	14.06	7000	0.4881	0.3657
0.069	15.06	7500	0.4957	0.3677
0.0679	16.06	8000	0.5241	0.3634
0.0618	17.07	8500	0.5091	0.3564
0.0576	18.07	9000	0.5055	0.3557
0.0493	19.08	9500	0.5013	0.3515
0.0469	20.08	10000	0.5506	0.3530
0.044	21.08	10500	0.5564	0.3528
0.0368	22.09	11000	0.5213	0.3509
0.0355	23.09	11500	0.5707	0.3495
0.0357	24.1	12000	0.5558	0.3483
0.0285	25.1	12500	0.5613	0.3455
0.0285	26.1	13000	0.5533	0.3480
0.0266	27.11	13500	0.5526	0.3462
0.0249	28.11	14000	0.5488	0.3429
0.0237	29.12	14500	0.5436	0.3401

Framework versions

• Transformers 4.17.0
• Pytorch 1.11.0+cu115
• Datasets 1.18.3
• Tokenizers 0.12.1