wav2vec2-base-timit-ms

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: 0.7589
• Wer: 0.3722

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: 80
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.0088	3.7	500	2.4873	1.0
1.0451	7.41	1000	0.9286	0.5470
0.4081	11.11	1500	0.5935	0.4397
0.2564	14.81	2000	0.6525	0.4292
0.183	18.52	2500	0.6578	0.4486
0.1481	22.22	3000	0.6786	0.4231
0.1299	25.93	3500	0.6660	0.4121
0.1044	29.63	4000	0.7713	0.4209
0.0953	33.33	4500	0.6728	0.4038
0.0746	37.04	5000	0.7122	0.4165
0.0627	40.74	5500	0.6950	0.4126
0.0554	44.44	6000	0.8237	0.4082
0.0494	48.15	6500	0.7311	0.3955
0.0426	51.85	7000	0.7717	0.3899
0.0368	55.56	7500	0.7490	0.3933
0.0315	59.26	8000	0.7056	0.3877
0.0274	62.96	8500	0.7897	0.3850
0.0237	66.67	9000	0.7715	0.3850
0.0223	70.37	9500	0.7774	0.3789
0.0177	74.07	10000	0.7598	0.3744
0.0182	77.78	10500	0.7589	0.3722

Framework versions

• Transformers 4.24.0
• Pytorch 2.0.0+cu118
• Datasets 1.18.3
• Tokenizers 0.13.3