alikanakar/whisper-synthesized-turkish-4-hour

Whisper Small Synthesized Turkish

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

• Loss: 0.2564
• Wer: 13.7267

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: 16
• 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: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.276	1.04	100	0.5859	92.8836
0.436	2.08	200	0.3916	19.5285
0.218	3.12	300	0.2345	13.2453
0.0903	4.17	400	0.2332	12.9737
0.0517	5.21	500	0.2360	14.3439
0.0302	6.25	600	0.2318	14.0415
0.0223	7.29	700	0.2372	13.9674
0.0085	8.33	800	0.2421	12.4738
0.0084	9.38	900	0.2424	12.3750
0.0043	10.42	1000	0.2421	12.8935
0.0034	11.46	1100	0.2478	13.6218
0.0025	12.5	1200	0.2490	14.7327
0.002	13.54	1300	0.2513	13.0910
0.0019	14.58	1400	0.2521	13.2453
0.0013	15.62	1500	0.2532	13.2144
0.0012	16.67	1600	0.2547	13.3132
0.001	17.71	1700	0.2552	13.7514
0.001	18.75	1800	0.2559	13.7452
0.001	19.79	1900	0.2563	13.7514
0.001	20.83	2000	0.2564	13.7267

Framework versions

• Transformers 4.28.0.dev0
• Pytorch 2.0.0+cu118
• Datasets 2.11.0
• Tokenizers 0.13.3