whisper-small-en-scratch-2

This model is a fine-tuned version of openai/whisper-small on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.5575
• Wer: 33.9105

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: 6
• eval_batch_size: 24
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 4
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.2222	2	1.8510	44.5887
No log	0.4444	4	1.8490	44.5887
0.7847	0.6667	6	1.8453	44.7330
0.7847	0.8889	8	1.8406	44.7330
0.7387	1.1111	10	1.8346	44.8773
0.7387	1.3333	12	1.8272	45.0216
0.7387	1.5556	14	1.8185	45.4545
0.7199	1.7778	16	1.8079	45.5988
0.7199	2.0	18	1.7952	45.5988
0.7153	2.2222	20	1.7810	45.1659
0.7153	2.4444	22	1.7643	45.0216
0.7153	2.6667	24	1.7443	34.3434
0.6205	2.8889	26	1.7225	34.1991
0.6205	3.1111	28	1.7001	34.1991
0.4817	3.3333	30	1.6743	33.6219
0.4817	3.5556	32	1.6421	33.6219
0.4817	3.7778	34	1.6030	33.9105
0.3973	4.0	36	1.5575	33.9105

Framework versions

• Transformers 4.42.0.dev0
• Pytorch 1.13.1+cu117
• Datasets 2.19.3.dev0
• Tokenizers 0.19.1