Mehtap/whisper-base-2023-01-31

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: 0.8800
• Wer: 0.8060
• Cer: 0.7585

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
1.8904	1.32	100	1.5873	0.8893	0.5437
0.8039	2.63	200	0.9239	0.9076	0.5721
0.5988	3.95	300	0.7970	0.7850	0.4821
0.384	5.26	400	0.7586	0.7164	0.5206
0.2643	6.58	500	0.7578	0.9130	0.6843
0.2026	7.89	600	0.7627	0.9147	0.7228
0.1091	9.21	700	0.8043	0.8363	0.8283
0.0623	10.53	800	0.8342	0.7615	0.7619
0.0436	11.84	900	0.8577	0.7079	0.6824
0.0348	13.16	1000	0.8800	0.8060	0.7585

Framework versions

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