wav2vec2-base-timit-finetune-3

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: 0.4387
• Wer: 0.2750

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: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 60
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.4692	6.94	500	0.9978	0.7091
0.3946	13.88	1000	0.3674	0.3253
0.1291	20.83	1500	0.3987	0.3042
0.074	27.77	2000	0.4292	0.2916
0.0487	34.72	2500	0.4302	0.2853
0.0368	41.66	3000	0.4222	0.2789
0.0281	48.61	3500	0.4481	0.2783
0.0237	55.55	4000	0.4387	0.2750

Framework versions

• Transformers 4.24.0
• Pytorch 1.12.1
• Datasets 2.7.0
• Tokenizers 0.11.0