ntviet/whisper-small-hre5.2

Whisper Small Hre 5.2, train/test split, ASR for male & female Hre voice

This model is a fine-tuned version of openai/whisper-small on the Hre audio dataset 8 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4342
• Cer Ortho: 19.1894
• Cer: 17.1849

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: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Cer Ortho	Cer
1.1862	0.9174	100	0.8429	36.8006	33.5051
0.3688	1.8349	200	0.5354	26.6679	24.2686
0.2010	2.7523	300	0.4394	24.7669	22.4655
0.1148	3.6697	400	0.3768	21.1980	19.2824
0.0712	4.5872	500	0.3922	22.0409	20.0368
0.0395	5.5046	600	0.4029	22.2561	19.8896
0.0281	6.4220	700	0.4333	22.5430	19.8344
0.0158	7.3394	800	0.3824	25.5918	23.5695
0.0167	8.2569	900	0.4228	20.6062	18.5465
0.0159	9.1743	1000	0.4062	20.4448	18.3257
0.0115	10.0917	1100	0.4105	23.8702	21.9687
0.0091	11.0092	1200	0.4257	24.3006	22.4655
0.0070	11.9266	1300	0.4210	21.0725	19.2456
0.0051	12.8440	1400	0.4185	19.6198	17.6633
0.0071	13.7615	1500	0.3942	19.9426	17.5529
0.0060	14.6789	1600	0.4541	21.0904	18.3993
0.0045	15.5963	1700	0.4144	19.1714	17.1481
0.0042	16.5138	1800	0.4424	19.9785	17.5345
0.0019	17.4312	1900	0.4293	19.4584	17.4057
0.0041	18.3486	2000	0.4049	18.5796	16.7065
0.0012	19.2661	2100	0.4143	18.6514	16.6329
0.0013	20.0	2180	0.4342	19.1894	17.1849

Framework versions

• Transformers 5.13.1
• Pytorch 2.11.0+cu128
• Datasets 2.18.0
• Tokenizers 0.22.2