arkanalexei/whisper-tiny-sundanese-pretrained-hanif

BisiX: Sundanese Whisper (Fine Tuned)

This model is a fine-tuned version of openai/whisper-tiny.en on the SU ID ASR dataset. It achieves the following results on the evaluation set:

• Loss: 0.1520
• Wer: 11.1640
• Cer: 5.3914

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: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 30
• training_steps: 270
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.1855	0.1765	30	1.2957	37.9416	13.6739
0.8989	0.3529	60	0.5241	24.9348	9.3570
0.3737	0.5294	90	0.3112	18.8135	6.9347
0.2528	0.7059	120	0.2354	13.5640	5.0195
0.1989	0.8824	150	0.2011	13.6989	7.9763
0.1554	1.0588	180	0.1727	10.2742	4.3784
0.1106	1.2353	210	0.1610	9.3393	3.5788
0.0918	1.4118	240	0.1560	12.3236	6.5779
0.0913	1.5882	270	0.1520	11.1640	5.3914

Framework versions

• Transformers 4.45.1
• Pytorch 2.4.1+cu124
• Datasets 3.0.1
• Tokenizers 0.20.0