Mehtap/base_04

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: 1.8804
• Wer: 2.0146
• Cer: 1.4030

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
1.4622	16.67	100	1.5376	0.8662	0.7357
0.297	33.33	200	1.2979	0.8675	0.6481
0.0163	50.0	300	1.5699	1.4066	1.0449
0.0034	66.67	400	1.6919	1.6416	1.1817
0.0017	83.33	500	1.7654	1.6943	1.2587
0.0011	100.0	600	1.8153	1.9908	1.4084
0.0008	116.67	700	1.8455	1.9817	1.3867
0.0007	133.33	800	1.8647	2.0479	1.4215
0.0006	150.0	900	1.8764	2.0489	1.4253
0.0006	166.67	1000	1.8804	2.0146	1.4030

Framework versions

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