wav2vec2-base-timit-eng

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: 0.4391
• Wer: 0.3836

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
3.8306	1.0	500	2.9588	1.0
2.1928	2.01	1000	1.2215	0.9355
1.1547	3.01	1500	0.9228	0.7135
0.9487	4.02	2000	0.7682	0.6513
0.8163	5.02	2500	0.7154	0.6164
0.6642	6.02	3000	0.6160	0.5919
0.6291	7.03	3500	0.6224	0.5485
0.601	8.03	4000	0.5927	0.5371
0.5443	9.04	4500	0.5757	0.5240
0.4798	10.04	5000	0.5673	0.5074
0.5142	11.04	5500	0.6138	0.5131
0.4044	12.05	6000	0.5899	0.5120
0.4214	13.05	6500	0.5443	0.4932
0.377	14.06	7000	0.6055	0.5337
0.3985	15.06	7500	0.5055	0.4812
0.3609	16.06	8000	0.5764	0.4600
0.299	17.07	8500	0.5524	0.4635
0.2984	18.07	9000	0.5272	0.4435
0.2908	19.08	9500	0.5393	0.4446
0.2714	20.08	10000	0.4548	0.4463
0.2285	21.08	10500	0.5126	0.4309
0.2245	22.09	11000	0.4770	0.4309
0.229	23.09	11500	0.4763	0.4150
0.2032	24.1	12000	0.5009	0.4127
0.2125	25.1	12500	0.4698	0.4087
0.1955	26.1	13000	0.4592	0.4001
0.1841	27.11	13500	0.4517	0.3898
0.164	28.11	14000	0.4628	0.3927
0.1687	29.12	14500	0.4391	0.3836

Framework versions

• Transformers 4.25.1
• Pytorch 1.13.0+cu116
• Datasets 1.18.3
• Tokenizers 0.13.2