Mehtap/base_09

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: 1.4238
• Wer: 0.9367
• Cer: 0.7611

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
1.6918	2.85	100	1.5023	0.7940	0.4289
0.6823	5.71	200	1.0475	0.8783	0.5573
0.4277	8.57	300	0.9944	0.8054	0.6120
0.2244	11.43	400	1.0460	0.6878	0.3825
0.1138	14.28	500	1.2059	0.7510	0.5020
0.0468	17.14	600	1.2180	1.1436	1.0719
0.0193	19.99	700	1.2801	1.1500	0.9344
0.0093	22.85	800	1.4574	0.9238	0.6799
0.0068	25.71	900	1.4137	0.9400	0.8128
0.0062	28.57	1000	1.4238	0.9367	0.7611

Framework versions

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