a-deji/whisper-small-hausa

Whisper Small Hausa Fine-Tuned

This model is a fine-tuned version of openai/whisper-small on the Publica AI Hausa Dataset (184 hours) dataset. It achieves the following results on the evaluation set:

• Loss: 0.1969
• Wer: 9.1731

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1644	3.0183	1000	0.2328	13.9821
0.0645	7.0092	2000	0.2068	12.2836
0.0665	4.8940	3000	0.1559	8.5712
0.0073	6.5253	4000	0.1765	8.9126
0.0013	8.1566	5000	0.1907	8.7752
0.0008	9.7879	6000	0.1969	9.1731

Framework versions

• Transformers 4.45.2
• Pytorch 2.9.1+cu128
• Datasets 4.4.1
• Tokenizers 0.20.3