wav2vec2-base

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: 3.0808
• Wer: 1.0

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: 4
• 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: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.7118	0.5	500	3.0635	1.0
2.9533	1.01	1000	3.0383	1.0
2.9493	1.51	1500	3.0638	1.0
2.9495	2.01	2000	3.0554	1.0
2.9468	2.51	2500	3.0630	1.0
2.9493	3.02	3000	3.0530	1.0
2.9457	3.52	3500	3.0534	1.0
2.9492	4.02	4000	3.0357	1.0
2.9444	4.52	4500	3.0366	1.0
2.9495	5.03	5000	3.0412	1.0
2.9468	5.53	5500	3.0331	1.0
2.9453	6.03	6000	3.0847	1.0
2.9484	6.53	6500	3.0661	1.0
2.9457	7.04	7000	3.0769	1.0
2.9449	7.54	7500	3.0701	1.0
2.9453	8.04	8000	3.1072	1.0
2.9436	8.54	8500	3.1043	1.0
2.9474	9.05	9000	3.0902	1.0
2.9452	9.55	9500	3.0879	1.0
2.9443	10.05	10000	3.1112	1.0
2.9436	10.55	10500	3.0946	1.0
2.9469	11.06	11000	3.0812	1.0
2.9434	11.56	11500	3.1112	1.0
2.9442	12.06	12000	3.0855	1.0
2.9436	12.56	12500	3.0786	1.0
2.9425	13.07	13000	3.0789	1.0
2.9418	13.57	13500	3.0786	1.0
2.9443	14.07	14000	3.0798	1.0
2.9449	14.57	14500	3.0808	1.0

Framework versions

• Transformers 4.11.3
• Pytorch 1.10.2
• Datasets 1.18.3
• Tokenizers 0.10.3