Whisper Small for Arabic ASR with diacritics

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

  • Loss: 0.188
  • Wer: 16.9

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: 8
  • 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: 4000

Training results

Training Loss Epoch Step Validation Loss Wer
0.0059 1.62 500 0.0259 18.8277
0.0019 3.24 1000 0.0223 17.1430
0.0007 4.85 1500 0.0211 17.0055
0.0003 6.47 2000 0.0198 16.4726
0.0 8.09 2500 0.0191 16.3351
0.0 9.71 3000 0.0187 16.3007
0.0 11.33 3500 0.0188 16.2491
0.0 12.94 4000 0.0188 16.9128

Framework versions

  • Transformers 4.40.0.dev0
  • Pytorch 2.1.2
  • Datasets 2.17.1
  • Tokenizers 0.15.2
Downloads last month
36
Safetensors
Model size
242M params
Tensor type
F32
·
Inference API
Unable to determine this model's library. Check the docs .

Model tree for moatazlumin/Arabic_ASR_whisper_small_with_diacritics

Finetuned
(2103)
this model

Evaluation results