wav2vec2-base-ami-fine-tuned

This model is a fine-tuned version of facebook/wav2vec2-large-lv60 on the EDINBURGHCSTR/AMI - IHM dataset. It achieves the following results on the evaluation set:

• Loss: 0.5988
• Wer: 0.3357

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: 16
• 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: 2.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.0732	0.1565	1000	1.1351	0.6738
1.4052	0.3131	2000	0.7311	0.4083
0.8798	0.4696	3000	0.5889	0.3604
0.4789	0.6262	4000	0.5681	0.3521
0.8011	0.7827	5000	0.5288	0.3382
1.4331	0.9393	6000	0.5386	0.3280
0.2201	1.0958	7000	0.5154	0.3198
0.1934	1.2523	8000	0.4895	0.3131
0.2713	1.4089	9000	0.4809	0.3065
0.1388	1.5654	10000	0.4984	0.3061
0.4085	1.7220	11000	0.4842	0.3082
0.3529	1.8785	12000	0.5417	0.3198

Framework versions

• Transformers 4.42.0.dev0
• Pytorch 2.3.0a0+gitcd033a1
• Datasets 2.19.1
• Tokenizers 0.19.1