Modelpublic

dodziraynard/Shona2

sourceHugging Faceapache-2.0updated 1y agoView on Hugging Face

1likes6downloads

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: 0.5275
—Wer Ortho: 46.3552
—Wer: 38.6876
—Cer: 13.2130

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
0.5021	0.4785	400	15.1787	0.5774	44.6759	52.4914
0.4833	0.9569	800	13.7387	0.5141	40.5820	48.5622
0.3765	1.4354	1200	13.0650	0.4926	38.5423	46.8196
0.3626	1.9139	1600	12.9516	0.4771	37.9238	46.1237
0.3109	2.3923	2000	12.3654	0.4750	37.0070	44.9041
0.3048	2.8708	2400	12.9748	0.4719	37.5137	45.5116
0.2446	3.3493	2800	0.4953	45.7020	37.8667	12.8493
0.2362	3.8278	3200	0.4882	45.9007	38.0896	13.0340
0.1642	4.3062	3600	0.5249	46.3910	38.3491	12.8627
0.1611	4.7847	4000	0.5275	46.3552	38.6876	13.2130

Framework versions

—Transformers 4.48.0
—Pytorch 2.6.0+cu124
—Datasets 4.0.0
—Tokenizers 0.21.2