sarpba/whisper-medium-cv18-hu-cleaned

Whisper medium Hu CV18

This model is a fine-tuned version of openai/whisper-medium on the Common Voice 18.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4280
• Wer Ortho: 26.6073
• Wer: 20.2222

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: 6.25e-06
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 500
• num_epochs: 3
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.195	0.2757	200	0.4397	29.9304	25.2390
0.1548	0.5513	400	0.4146	28.6874	23.6903
0.126	0.8270	600	0.4022	28.3332	22.4632
0.077	1.1027	800	0.4045	27.5831	21.3673
0.0744	1.3784	1000	0.4096	27.8566	21.4102
0.0718	1.6540	1200	0.3955	26.5619	20.7733
0.0681	1.9297	1400	0.3990	26.5267	20.6207
0.032	2.2054	1600	0.4056	25.8913	20.1680
0.0323	2.4810	1800	0.4232	26.1182	20.2878
0.0356	2.7567	2000	0.4280	26.6073	20.2222

Framework versions

• Transformers 4.44.0
• Pytorch 2.3.0+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1