wav2vec2-15

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: 0.8623
• Wer: 0.8585

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: 1e-05
• 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: 400
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
9.6808	1.37	200	3.7154	1.0
3.0784	2.74	400	3.1542	1.0
2.8919	4.11	600	2.9918	1.0
2.8317	5.48	800	2.8971	1.0
2.7958	6.85	1000	2.8409	1.0
2.7699	8.22	1200	2.8278	1.0
2.6365	9.59	1400	2.4657	1.0
2.1096	10.96	1600	1.8358	0.9988
1.6485	12.33	1800	1.4525	0.9847
1.3967	13.7	2000	1.2467	0.9532
1.2492	15.07	2200	1.1261	0.9376
1.1543	16.44	2400	1.0654	0.9194
1.0863	17.81	2600	1.0136	0.9161
1.0275	19.18	2800	0.9601	0.8827
0.9854	20.55	3000	0.9435	0.8878
0.9528	21.92	3200	0.9170	0.8807
0.926	23.29	3400	0.9121	0.8783
0.9025	24.66	3600	0.8884	0.8646
0.8909	26.03	3800	0.8836	0.8690
0.8717	27.4	4000	0.8810	0.8646
0.8661	28.77	4200	0.8623	0.8585

Framework versions

• Transformers 4.19.2
• Pytorch 1.11.0+cu113
• Datasets 2.2.2
• Tokenizers 0.12.1