Mehtap/base_02

base Turkish Whisper (bTW)

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

• Loss: 2.1975
• Wer: 1.6817
• Cer: 1.2800

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: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.5514	33.31	100	1.6389	0.8196	0.8754
0.1703	66.62	200	1.6896	1.0058	0.6987
0.0039	99.92	300	1.9380	1.7011	1.1631
0.0015	133.31	400	2.0324	1.6950	1.2498
0.0008	166.62	500	2.0957	1.4898	1.0992
0.0005	199.92	600	2.1417	1.7320	1.2528
0.0004	233.31	700	2.1681	1.6077	1.1845
0.0003	266.62	800	2.1847	1.625	1.2008
0.0003	299.92	900	2.1944	1.6515	1.2196
0.0003	333.31	1000	2.1975	1.6817	1.2800

Framework versions

• Transformers 4.26.0
• Pytorch 1.12.0+cu102
• Datasets 2.9.0
• Tokenizers 0.13.2