whisper-NST2

This model is a fine-tuned version of openai/whisper-small on the NBAILAB/NST - NO-CLOSE dataset. It achieves the following results on the evaluation set:

• Loss: 0.2990
• Wer: 7.7537

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: 4e-05
• train_batch_size: 96
• 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
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1846	0.1	1000	0.3460	14.9373
0.1325	0.2	2000	0.3413	11.4025
0.1135	0.3	3000	0.3428	12.6568
0.0955	0.4	4000	0.3140	10.7184
0.0871	0.5	5000	0.2907	9.4641
0.0774	0.6	6000	0.3019	11.4025
0.041	1.1	7000	0.2897	9.0080
0.0306	1.2	8000	0.3013	7.6397
0.0279	1.3	9000	0.2958	9.1220
0.0239	1.4	10000	0.2990	7.7537

Framework versions

• Transformers 4.25.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.6.1
• Tokenizers 0.13.1