wav2vec2-23

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: 3.1230
• 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: 0.0003
• 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
4.2642	1.37	200	2.9756	1.0
2.8574	2.74	400	3.1631	1.0
2.8588	4.11	600	3.1208	1.0
2.8613	5.48	800	3.1113	1.0
2.8599	6.85	1000	3.2679	1.0
2.8577	8.22	1200	3.0904	1.0
2.8575	9.59	1400	3.2444	1.0
2.8538	10.96	1600	3.0674	1.0
2.8564	12.33	1800	3.1957	1.0
2.8555	13.7	2000	3.0881	1.0
2.8542	15.07	2200	3.1488	1.0
2.8538	16.44	2400	3.1184	1.0
2.854	17.81	2600	3.1133	1.0
2.8553	19.18	2800	3.1508	1.0
2.8534	20.55	3000	3.0646	1.0
2.8538	21.92	3200	3.1374	1.0
2.8545	23.29	3400	3.1020	1.0
2.8539	24.66	3600	3.1631	1.0
2.8558	26.03	3800	3.1063	1.0
2.8508	27.4	4000	3.1271	1.0
2.8537	28.77	4200	3.1230	1.0

Framework versions

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