Mehtap/base_05

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: 1.5006
• Wer: 1.3698
• Cer: 1.1255

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
1.8141	5.53	100	1.4784	0.7680	0.4463
0.673	11.11	200	1.0561	0.8175	0.5889
0.2762	16.64	300	1.0746	0.8564	0.5887
0.0852	22.21	400	1.2061	1.4290	0.9567
0.0199	27.75	500	1.2649	1.0706	0.9168
0.0087	33.32	600	1.4641	1.2417	1.0328
0.0041	38.85	700	1.4685	1.2806	0.9546
0.003	44.43	800	1.4830	1.3633	1.0236
0.0026	49.96	900	1.4964	1.3698	1.0375
0.0025	55.53	1000	1.5006	1.3698	1.1255

Framework versions

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