whisper-atco2-medium

This model is a fine-tuned version of openai/whisper-medium on the luigisaetta/atco2_normalized_augmented dataset. It achieves the following results on the evaluation set:

• Loss: 0.6129
• Wer: 17.5052

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: 1e-05
• train_batch_size: 2
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• training_steps: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.3939	1.06	50	1.8493	66.5618
0.5127	2.13	100	0.5119	30.6080
0.0626	3.19	150	0.5410	20.4403
0.0157	4.25	200	0.5775	19.8113
0.0107	5.32	250	0.5552	19.7065
0.0044	6.38	300	0.5723	18.1342
0.0013	7.45	350	0.5763	17.7149
0.0005	8.51	400	0.6053	17.7149
0.0004	9.57	450	0.6109	17.5052
0.0004	10.64	500	0.6129	17.5052

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.11.0