Mehtap/base_08

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: 1.1451
• Wer: 1.0165
• Cer: 0.7894

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
1.6901	4.54	100	1.3928	0.8093	0.4264
0.6163	9.09	200	0.8885	0.7907	0.4532
0.2692	13.63	300	0.8719	0.7823	0.4474
0.1148	18.18	400	0.9275	0.7393	0.4280
0.04	22.72	500	1.0308	0.8162	0.5241
0.0114	27.27	600	1.0885	0.9666	0.7902
0.0051	31.81	700	1.1159	0.9594	0.6967
0.0036	36.36	800	1.1301	1.0451	0.7819
0.0031	40.9	900	1.1415	1.0496	0.8072
0.0028	45.45	1000	1.1451	1.0165	0.7894

Framework versions

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