whisper-small-noisy-hi

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

• Loss: 1.5460
• Wer: 74.5720

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: 48
• 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: 500
• training_steps: 3000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.5752	0.46	50	2.2665	120.7418
1.6855	0.92	100	1.6174	92.1494
1.4464	1.38	150	1.4430	92.0543
1.3211	1.83	200	1.3179	88.5094
1.1732	2.29	250	1.2025	86.2182
1.0507	2.75	300	1.0736	83.7628
0.8575	3.21	350	0.9902	80.8404
0.8096	3.67	400	0.9516	80.1833
0.7257	4.13	450	0.9286	78.7740
0.6689	4.59	500	0.9091	77.0621
0.6331	5.05	550	0.9014	76.5087
0.5123	5.5	600	0.9030	74.3213
0.505	5.96	650	0.8833	76.0851
0.3716	6.42	700	0.9274	75.5144
0.3759	6.88	750	0.9227	74.1657
0.2658	7.34	800	0.9754	77.3993
0.2624	7.8	850	0.9800	74.9784
0.1755	8.26	900	1.0364	74.5807
0.1771	8.72	950	1.0549	76.0678
0.1239	9.17	1000	1.1081	74.8314
0.112	9.63	1050	1.1373	74.9524
0.0942	10.09	1100	1.1697	75.2205
0.0691	10.55	1150	1.2068	76.6384
0.0659	11.01	1200	1.2280	75.6095
0.0417	11.47	1250	1.2840	74.9697
0.0416	11.93	1300	1.3025	75.9035
0.025	12.39	1350	1.3342	76.1110
0.0258	12.84	1400	1.3580	74.9438
0.0182	13.3	1450	1.4077	75.9467
0.0154	13.76	1500	1.4214	75.1167
0.0131	14.22	1550	1.4525	74.8660
0.0119	14.68	1600	1.4903	74.7709
0.011	15.14	1650	1.5147	75.0476
0.0079	15.6	1700	1.5375	75.9727
0.0087	16.06	1750	1.5460	74.5720

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 1.12.1
• Datasets 2.16.1
• Tokenizers 0.15.0