Modelpublic

dodziraynard/Shona

sourceHugging Faceapache-2.0updated 1y agoView on Hugging Face

1likes8downloads

Model Card

UG Speech Data ASR - Ewe nornmaliser

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

—Loss: 0.5273
—Wer Ortho: 46.0461
—Wer: 38.3491
—Cer: 13.0384

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
—trainbatchsize: 16
—evalbatchsize: 16
—seed: 42
—optimizer: Use OptimizerNames.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: constantwithwarmup
—lrschedulerwarmup_steps: 50
—training_steps: 4000
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer	Wer Ortho
0.5022	0.4785	400	15.0475	0.5773	44.5732	52.3734
0.4835	0.9569	800	13.6924	0.5142	40.5166	48.4899
0.3764	1.4354	1200	13.2187	0.4926	38.7241	47.1020
0.3624	1.9139	1600	12.8324	0.4770	37.8553	46.0811
0.3165	2.3923	2000	0.4770	45.1081	37.1660	12.5025
0.3058	2.8708	2400	0.4728	45.5634	37.5822	12.8574
0.2386	3.3493	2800	0.4945	45.8291	38.0272	12.8462
0.2334	3.8278	3200	0.4874	45.7743	38.0440	12.8868
0.1662	4.3062	3600	0.5242	46.6003	38.5020	12.9679
0.1615	4.7847	4000	0.5273	46.0461	38.3491	13.0384

Framework versions

—Transformers 4.48.0
—Pytorch 2.6.0+cu124
—Datasets 4.0.0
—Tokenizers 0.21.4