whisper-small-en

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

• Loss: 6.7832
• Wer: 124.5115

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: 0.0005
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-06
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2
• training_steps: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
9.6259	1.57	5	10.7408	1127.3535
11.5288	3.29	10	9.2534	100.0
10.9249	4.86	15	7.8357	100.0
7.0442	6.57	20	6.9971	595.3819
8.6762	8.29	25	5.6135	312.2558
5.4239	9.86	30	5.4885	97.1581
4.986	11.57	35	5.2888	628.7744
6.708	13.29	40	4.9665	277.6199
3.9096	14.86	45	5.0861	631.9716
3.2326	16.57	50	5.0090	279.7513
3.9691	18.29	55	5.0804	133.2149
1.8661	19.86	60	5.4423	317.5844
1.1588	21.57	65	5.7955	119.5382
1.0355	23.29	70	6.0458	190.2309
0.3455	24.86	75	6.3057	106.7496
0.142	26.57	80	6.5767	209.9467
0.1722	28.29	85	6.5937	101.4210
0.0816	29.86	90	6.7679	149.7336
0.079	31.57	95	6.8008	133.5702
0.1007	33.29	100	6.7832	124.5115

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.12.1+cu113
• Datasets 2.7.1
• Tokenizers 0.13.2