whisper-synthesized-turkish-2-hour

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: 0.3457
• Wer: 20.3461

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
1.3076	2.08	100	0.6286	96.2530
0.421	4.17	200	0.4269	22.7088
0.1585	6.25	300	0.2911	22.3150
0.0482	8.33	400	0.3047	16.6706
0.022	10.42	500	0.3086	16.5752
0.013	12.5	600	0.3209	19.7613
0.0049	14.58	700	0.3185	16.1575
0.0025	16.67	800	0.3278	16.7303
0.0019	18.75	900	0.3239	20.5012
0.0019	20.83	1000	0.3307	19.7613
0.0011	22.92	1100	0.3329	20.5728
0.0008	25.0	1200	0.3361	20.5609
0.0007	27.08	1300	0.3383	20.3341
0.0006	29.17	1400	0.3403	20.2029
0.0006	31.25	1500	0.3418	20.3699
0.0006	33.33	1600	0.3432	20.0477
0.0005	35.42	1700	0.3442	20.0835
0.0005	37.5	1800	0.3450	20.1313
0.0005	39.58	1900	0.3454	20.3699
0.0005	41.67	2000	0.3457	20.3461

Framework versions

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