Whisper Tiny ar

This model is a fine-tuned version of openai/whisper-tiny on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:

Loss: 0.5962
Wer: 52.3622
Cer: 18.7245

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: 64
eval_batch_size: 64
seed: 42
optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.04
training_steps: 18000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.1234	0.0556	1000	0.8106	67.0994	25.5658
0.8957	0.1111	2000	0.7234	62.5310	23.3911
0.7114	0.1667	3000	0.6871	59.7675	21.8722
0.6346	0.2222	4000	0.6637	58.1976	20.7336
0.4961	0.2778	5000	0.6545	57.7404	20.7048
0.4354	0.3333	6000	0.6473	56.7948	20.2061
0.3924	0.3889	7000	0.6325	55.8400	20.0139
0.3466	0.4444	8000	0.6274	55.4176	20.1441
0.2979	0.5	9000	0.6206	54.6997	19.6005
0.3099	0.5556	10000	0.6150	54.0166	19.3231
0.2681	0.6111	11000	0.6120	53.5980	19.1106
0.2383	0.6667	12000	0.6113	53.5576	19.4238
0.2582	0.7222	13000	0.6060	52.7515	18.7573
0.1543	0.7778	14000	0.6018	52.6175	18.4895
0.2356	0.8333	15000	0.6023	52.9902	18.9782
0.2031	0.8889	16000	0.5984	52.5165	18.8550
0.2437	0.9444	17000	0.5951	52.4926	18.7514
0.2269	1.0	18000	0.5962	52.3622	18.7245

Framework versions

Transformers 4.48.0.dev0
Pytorch 2.5.1+cu121
Datasets 3.6.0
Tokenizers 0.21.0

Citation

Please cite the model using the following BibTeX entry:

bibtex

@misc{deepdml/whisper-tiny-ar-mix-norm,
      title={Fine-tuned Whisper tiny ASR model for speech recognition in Arabic},
      author={Jimenez, David},
      howpublished={\url{https://huggingface.co/deepdml/whisper-tiny-ar-mix-norm}},
      year={2026}
    }