whisper-tiny.en-finetuned

This model is a fine-tuned version of openai/whisper-tiny.en on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.4890
• Wer: 0.8750
• Cer: 0.4454
• Ser: 0.055

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-06
• train_batch_size: 32
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 20
• training_steps: 220
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Ser
6.4419	0.4	10	6.4480	45.25	16.3573	1.0
6.2706	0.8	20	5.9268	39.375	14.1797	1.0
5.3444	1.2	30	4.8418	19.875	7.3002	1.0
4.3269	1.6	40	3.9103	9.375	3.8357	1.0
3.6103	2.0	50	3.3228	7.0000	3.2170	1.0
3.0331	2.4	60	2.7725	5.375	2.3509	1.0
2.5529	2.8	70	2.3913	4.75	1.8312	1.0
2.2359	3.2	80	2.1217	3.375	1.4353	1.0
1.9777	3.6	90	1.8790	3.0	1.3116	1.0
1.7863	4.0	100	1.7402	2.875	1.0393	0.98
1.6854	4.4	110	1.6736	2.625	0.9651	0.925
1.6297	4.8	120	1.6290	2.625	0.9404	0.765
1.5974	5.2	130	1.5959	2.5	0.9651	0.57
1.5673	5.6	140	1.5702	1.5	0.6434	0.385
1.5397	6.0	150	1.5497	1.25	0.5692	0.27
1.524	6.4	160	1.5333	1.125	0.5444	0.22
1.509	6.8	170	1.5199	0.8750	0.4454	0.155
1.4986	7.2	180	1.5091	0.8750	0.4454	0.095
1.4872	7.6	190	1.5011	0.8750	0.4454	0.075
1.4844	8.0	200	1.4950	0.8750	0.4454	0.075
1.4743	8.4	210	1.4910	0.8750	0.4454	0.065
1.4746	8.8	220	1.4890	0.8750	0.4454	0.055

Framework versions

• Transformers 4.39.3
• Pytorch 2.2.2+cu121
• Datasets 2.14.5
• Tokenizers 0.15.2