xklzv/whisper-small-dv

Whisper Small Dv - Sanchit Gandhi

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

• Loss: 0.1480
• Wer Ortho: 42.2207
• Wer: 7.0895

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: 16
• 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: constant_with_warmup
• lr_scheduler_warmup_steps: 50
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.0605	2.3923	500	0.0790	46.1564	7.7177
0.0281	4.7847	1000	0.0854	43.5270	7.3792
0.0085	7.1770	1500	0.1165	43.3261	7.2649
0.0051	9.5694	2000	0.1230	43.4601	7.0120
0.0031	11.9617	2500	0.1358	42.3045	6.8937
0.0025	14.3541	3000	0.1438	42.9744	6.9957
0.0035	16.7464	3500	0.1413	42.3547	6.8040
0.0017	19.1388	4000	0.1480	42.2207	7.0895

Framework versions

• Transformers 4.48.3
• Pytorch 2.10.0+cu130
• Datasets 4.5.0
• Tokenizers 0.21.4