clt013/whisper-medium-ft-malay

Whisper Medium FT Malay - CLT013

This model is a fine-tuned version of openai/whisper-medium on the Malay Speech 1.6 million dataset. It achieves the following results on the evaluation set:

• Loss: 0.7057
• Wer: 39.6567

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: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.0434	0.1	100	0.9250	53.3417
0.8131	0.2	200	0.8394	46.5908
0.7852	0.3	300	0.8033	45.1635
0.7643	0.4	400	0.7769	53.5732
0.7424	0.5	500	0.7582	46.6969
0.7406	0.6	600	0.7451	39.6760
0.7913	0.7	700	0.7288	39.3866
0.7452	0.8	800	0.7164	37.9979
0.718	0.9	900	0.7099	38.7694
0.7328	1.0	1000	0.7057	39.6567

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1