whisper-small-ne-NP

This model is a fine-tuned version of openai/whisper-small on the common_voice_13_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.6005
• Wer: 57.3876

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: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.9935	0.17	100	1.3460	91.4347
0.6624	0.35	200	1.0307	85.6531
0.5002	0.52	300	0.8406	77.5161
0.4426	0.7	400	0.7038	76.2313
0.3063	0.87	500	0.5308	71.5203
0.1949	1.05	600	0.5200	66.1670
0.1974	1.22	700	0.5140	65.0964
0.1734	1.4	800	0.4423	67.6660
0.1619	1.57	900	0.4705	62.0985
0.1697	1.75	1000	0.4676	67.0236
0.1536	1.92	1100	0.4441	62.7409
0.0722	2.1	1200	0.4492	58.0300
0.0674	2.27	1300	0.4597	59.9572
0.0766	2.45	1400	0.4720	62.3126
0.0732	2.62	1500	0.4720	60.5996
0.0737	2.8	1600	0.4704	61.0278
0.0833	2.97	1700	0.4711	59.7430
0.0421	3.15	1800	0.5040	60.5996
0.0444	3.32	1900	0.5096	62.5268
0.0343	3.5	2000	0.5276	62.5268
0.0347	3.67	2100	0.5068	57.3876
0.0326	3.85	2200	0.5143	59.3148
0.0219	4.02	2300	0.5225	59.3148
0.0129	4.2	2400	0.5353	59.1006
0.0159	4.37	2500	0.5639	56.9593
0.0168	4.55	2600	0.5303	55.8887
0.0131	4.72	2700	0.5455	58.6724
0.0122	4.9	2800	0.5548	56.5310
0.0035	5.07	2900	0.5661	56.7452
0.0027	5.24	3000	0.5789	57.6017
0.0034	5.42	3100	0.5887	59.1006
0.0047	5.59	3200	0.5853	59.9572
0.0054	5.77	3300	0.5912	58.4582
0.0042	5.94	3400	0.5862	59.3148
0.0013	6.12	3500	0.5935	56.7452
0.001	6.29	3600	0.5991	57.3876
0.0008	6.47	3700	0.6012	57.6017
0.0014	6.64	3800	0.6002	57.8158
0.001	6.82	3900	0.6006	57.8158
0.0013	6.99	4000	0.6005	57.3876

Framework versions

• Transformers 4.28.1
• Pytorch 2.0.0
• Datasets 2.11.0
• Tokenizers 0.13.3