wav2vec2-base-librispeech-demo-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: 0.2489
• Wer: 0.1673

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: 32
• 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
3.7801	2.2	500	2.1946	1.0068
0.7656	4.41	1000	0.3422	0.3137
0.3193	6.61	1500	0.2528	0.2327
0.2202	8.81	2000	0.2141	0.2121
0.1703	11.01	2500	0.2121	0.1966
0.1388	13.22	3000	0.3337	0.2325
0.119	15.42	3500	0.2342	0.1847
0.0992	17.62	4000	0.2356	0.1785
0.0875	19.82	4500	0.2534	0.1810
0.0769	22.03	5000	0.2491	0.1765
0.0661	24.23	5500	0.2513	0.1710
0.0587	26.43	6000	0.2546	0.1686
0.0544	28.63	6500	0.2489	0.1673

Framework versions

• Transformers 4.29.2
• Pytorch 2.0.1+cu118
• Datasets 2.12.0
• Tokenizers 0.13.3