wav2vecvanilla

This model is a fine-tuned version of facebook/wav2vec2-base-960h on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.8028
• Wer: 0.3041

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: 16
• 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: 500
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.1729	1.69	100	1.0048	0.3681
1.0722	3.39	200	0.9117	0.3511
0.9994	5.08	300	0.9302	0.3436
0.9576	6.78	400	0.8246	0.3320
0.9826	8.47	500	0.7846	0.3343
0.801	10.17	600	0.8600	0.3269
0.8174	11.86	700	0.7871	0.3186
0.7162	13.56	800	0.8164	0.3186
0.7447	15.25	900	0.8965	0.3084
0.6889	16.95	1000	0.8239	0.3057
0.6739	18.64	1100	0.8028	0.3041

Framework versions

• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2