abbenedekwhisper-tiny.en-finetuned-D3K

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: 1.7482
• Cer: 20.4786
• Wer: 36.7550
• Ser: 32.4561

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: 40
• training_steps: 400
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer	Ser
6.5477	0.21	20	6.1679	28.8815	65.5629	47.3684
5.8649	0.43	40	5.1645	26.6555	59.6026	42.1053
4.4955	0.64	60	3.9916	28.6032	57.6159	40.3509
3.2461	0.85	80	3.0695	25.2087	49.0066	35.0877
2.464	1.06	100	2.4087	22.6489	42.0530	33.3333
2.0699	1.28	120	2.1494	22.7045	40.0662	32.4561
1.9909	1.49	140	2.0336	22.5932	39.0728	32.4561
1.8902	1.7	160	1.9646	21.5915	38.0795	33.3333
1.8438	1.91	180	1.9208	23.0384	38.7417	34.2105
1.7684	2.13	200	1.8850	22.2037	38.4106	34.2105
1.729	2.34	220	1.8487	21.9254	37.4172	33.3333
1.6901	2.55	240	1.8251	21.9254	37.0861	32.4561
1.7107	2.77	260	1.8077	21.9811	37.7483	33.3333
1.665	2.98	280	1.7931	22.0924	39.4040	34.2105
1.6214	3.19	300	1.7799	22.0924	39.4040	34.2105
1.6124	3.4	320	1.7692	21.2020	37.4172	33.3333
1.6064	3.62	340	1.7602	21.2020	37.4172	33.3333
1.6154	3.83	360	1.7538	20.4786	36.7550	32.4561
1.6167	4.04	380	1.7497	20.4786	36.7550	32.4561
1.5972	4.26	400	1.7482	20.4786	36.7550	32.4561

Framework versions

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