whisper-small-clearglobal-hausa-asr-1.0.0

This model is a fine-tuned version of openai/whisper-small on the TWB Voice 1.0 dataset.

It achieves the following results on the internal evaluation set:

Training and evaluation data

This model was trained by colleagues from the Makerere University Centre for Artificial Intelligence and Data Science in collaboration with CLEAR Global. We gratefully acknowledge their expertise and partnership.

Model was trained and tested on the approved Hausa subset of TWB Voice 1.0 dataset.

Train/dev/test portions correspond to the splits in this dataset version. Test splits consist of speakers not present in train and dev splits.

We also tested on external datasets: Common voice v17 and Naija Voices test splits.

The evaluation results are as follows:

The following hyperparameters were used during training:

learning_rate: 0.0001
train_batch_size: 32
eval_batch_size: 32
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 64
optimizer: Use OptimizerNames.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: 0.08
num_epochs: 50.0
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.8668	1.0	423	0.3216	0.2559	0.0794
0.2265	2.0	846	0.2439	0.1971	0.0668
0.1419	3.0	1269	0.2218	0.1832	0.0721
0.1109	4.0	1692	0.2233	0.1639	0.0639
0.088	5.0	2115	0.2168	0.1488	0.0578
0.0591	6.0	2538	0.1909	0.1358	0.0548
0.0438	7.0	2961	0.1867	0.1165	0.0488
0.033	8.0	3384	0.1855	0.1190	0.0510
0.028	9.0	3807	0.1898	0.1177	0.0525
0.0223	10.0	4230	0.1833	0.1139	0.0522
0.0182	11.0	4653	0.1835	0.1090	0.0469
0.0148	12.0	5076	0.1771	0.1111	0.0517
0.0117	13.0	5499	0.1726	0.1047	0.0479
0.0112	14.0	5922	0.1698	0.0942	0.0425
0.0102	15.0	6345	0.1691	0.0940	0.0413
0.0099	16.0	6768	0.1701	0.0960	0.0456
0.0093	17.0	7191	0.1793	0.1000	0.0456
0.0082	18.0	7614	0.1724	0.0957	0.0443
0.0065	19.0	8037	0.1726	0.0989	0.0465

Safetensors

Model size

0.2B params

Tensor type

F32

Base model

Finetuned

(3164)

this model