clt013/whisper-small-ft-malay-test-3

---

Whisper Large v3 FT Malay - CLT013

This model is a fine-tuned version of openai/whisper-small on the Malay Speech 3k dataset. It achieves the following results on the evaluation set:

• Loss: 0.545344
• Wer: 27.169943

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
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7275	0.1	100	0.677592	38.9111
0.521	0.2	200	0.565486	36.6151
0.3128	0.3	300	0.525294	29.7965
0.2964	0.4	400	0.500519	35.2235
0.1631	0.5	500	0.508256	36.2845
0.0731	0.6	600	0.532225	38.4414
0.0548	0.7	700	0.519905	27.2743
0.0289	0.8	800	0.533013	27.6917
0.0131	0.9	900	0.548259	26.9090
0.0071	1.0	1000	0.545344	27.1699

Framework versions

• Transformers 4.41.2
• Datasets 2.19.1
• Tokenizers 0.19.1