lorenzoncina/whisper-small-ar

Whisper Small Arabic

This model is a fine-tuned version of openai/whisper-small on the mozilla-foundation/common_voice_11_0 ar dataset. It achieves the following results on the evaluation set:

• Loss: 0.4948
• Wer: 54.08

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: 32
• eval_batch_size: 2
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1885	1.03	1000	0.3950	66.44
0.0794	3.0	2000	0.3950	58.5507
0.0286	4.04	3000	0.4602	63.88
0.0128	6.01	4000	0.4948	54.08
0.0048	7.04	5000	0.5466	57.9867
0.0029	9.01	6000	0.5710	55.4147
0.0013	10.05	7000	0.5996	58.7707
0.0008	12.02	8000	0.6179	54.748
0.0006	13.05	9000	0.6343	56.2613
0.0003	15.02	10000	0.6388	56.228

Framework versions

• Transformers 4.28.0.dev0
• Pytorch 2.0.0+cu117
• Datasets 2.11.1.dev0
• Tokenizers 0.13.2