octava/whisper-small-indonesian-disaster-vanilla

Whisper Medium Indonesian for Disaster Response

This model is a fine-tuned version of openai/whisper-small on the Indonesian Speech Dataset (InaVoCript, Fleurs, OpenSLR Javanese) dataset. It achieves the following results on the evaluation set:

• Loss: 0.4044
• Wer: 16.2338

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
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 5000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0136	9.9010	500	0.3101	16.2013
0.0007	19.8020	1000	0.3472	15.4870
0.0004	29.7030	1500	0.3614	15.5844
0.0002	39.6040	2000	0.3725	15.7468
0.0002	49.5050	2500	0.3812	15.7468
0.0001	59.4059	3000	0.3891	15.7468
0.0001	69.3069	3500	0.3957	16.0390
0.0001	79.2079	4000	0.4005	16.0390
0.0001	89.1089	4500	0.4037	16.1364
0.0001	99.0099	5000	0.4044	16.2338

Framework versions

• Transformers 4.45.0
• Pytorch 2.6.0+cu124
• Datasets 3.0.1
• Tokenizers 0.20.0