ToeLay/whisper_large_v3_turbo_mm

Whisper Large V3 Turbo Burmese Finetune

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the Myanmar Speech Dataset (OpenSLR-80) dataset. It achieves the following results on the evaluation set:

• Loss: 0.2310
• Wer: 55.7881

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: 0.0003
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.2
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7755	1.0	143	0.3657	92.8317
0.2954	2.0	286	0.2669	85.6189
0.2483	3.0	429	0.2830	82.7248
0.2332	4.0	572	0.2922	83.3927
0.204	5.0	715	0.2338	78.8068
0.1612	6.0	858	0.1876	74.8442
0.1203	7.0	1001	0.1940	72.1728
0.0919	8.0	1144	0.1639	65.8504
0.0663	9.0	1287	0.1610	62.5557
0.0461	10.0	1430	0.1633	63.2235
0.0336	11.0	1573	0.1830	62.8228
0.0238	12.0	1716	0.1777	60.5521
0.0153	13.0	1859	0.1783	59.4835
0.0099	14.0	2002	0.1945	58.2369
0.0066	15.0	2145	0.2002	57.1683
0.003	16.0	2288	0.2148	57.1683
0.0015	17.0	2431	0.2241	55.9662
0.0006	18.0	2574	0.2286	56.2778
0.0003	19.0	2717	0.2296	55.8771
0.0001	20.0	2860	0.2310	55.7881

Framework versions

• Transformers 4.46.2
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.20.3