Mehtap/whisper-base

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: 0.0009
• Wer: 0.0
• Cer: 0.0

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• 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
1.8786	6.63	100	1.3510	0.7866	0.6649
0.4559	13.32	200	0.3395	0.3590	0.2157
0.0793	19.95	300	0.0564	0.0996	0.0531
0.0137	26.63	400	0.0120	0.0017	0.0017
0.0042	33.32	500	0.0032	0.0	0.0
0.0021	39.95	600	0.0018	0.0	0.0
0.0014	46.63	700	0.0013	0.0	0.0
0.0012	53.32	800	0.0011	0.0	0.0
0.001	59.95	900	0.0010	0.0	0.0
0.001	66.63	1000	0.0009	0.0	0.0

Framework versions

• Transformers 4.25.1
• Pytorch 1.9.1+cu111
• Datasets 2.7.1
• Tokenizers 0.13.2