OwLim/whisper-sundanese-finetune

Whisper Small Sundanese

This model is a fine-tuned version of openai/whisper-small on the SLR44 Augmented Sundanese dataset. It achieves the following results on the evaluation set:

• Loss: 0.1282
• Wer: 13.0988

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: 16
• eval_batch_size: 8
• seed: 42
• 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_steps: 100
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.6373	0.4292	100	0.3069	25.1403
0.4107	0.8584	200	0.1883	17.6953
0.1919	1.2876	300	0.1628	15.6884
0.1769	1.7167	400	0.1486	14.1562
0.0698	2.1459	500	0.1371	14.2641
0.0599	2.5751	600	0.1334	13.6167
0.0604	3.0043	700	0.1288	13.5304
0.0259	3.4335	800	0.1300	13.1636
0.025	3.8627	900	0.1280	13.1420
0.0169	4.2918	1000	0.1282	13.0988

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1