Mehtap/base_11

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: 1.0034
• Wer: 0.9507
• Cer: 0.9543

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
1.6746	2.63	100	1.4311	0.8342	0.5210
0.7117	5.26	200	0.8645	0.9008	0.5476
0.4373	7.89	300	0.7748	0.7412	0.5489
0.2419	10.53	400	0.7788	0.6967	0.4042
0.1359	13.16	500	0.8320	0.6912	0.5735
0.055	15.79	600	0.8891	0.7571	0.7292
0.0268	18.42	700	0.9250	0.7480	0.6051
0.0133	21.05	800	0.9747	0.6906	0.7730
0.0088	23.68	900	0.9968	0.8349	0.8106
0.0077	26.32	1000	1.0034	0.9507	0.9543

Framework versions

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