Whisper Medium PL

This model is a fine-tuned version of openai/whisper-medium on the Common Voice 11.0 and the FLEURS datasets. It achieves the following results on the evaluation set:

Loss: 0.3947
Wer: 8.6872

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: 4
eval_batch_size: 1
seed: 42
gradient_accumulation_steps: 8
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: 500
training_steps: 8000
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0805	0.48	500	0.2556	10.4888
0.0685	0.96	1000	0.2462	10.7608
0.0356	1.45	1500	0.2561	9.6728
0.0337	1.93	2000	0.2327	9.6459
0.017	2.41	2500	0.2444	9.9464
0.0179	2.9	3000	0.2554	9.6476
0.0056	3.38	3500	0.3001	9.3638
0.007	3.86	4000	0.2809	9.2245
0.0033	4.34	4500	0.3235	9.3437
0.0024	4.83	5000	0.3148	9.0633
0.0008	5.31	5500	0.3416	9.0112
0.0011	5.79	6000	0.3876	9.1858
0.0004	6.27	6500	0.3745	8.7292
0.0003	6.76	7000	0.3704	9.0314
0.0003	7.24	7500	0.3929	8.6553
0.0002	7.72	8000	0.3947	8.6872