nonhmello/whisper-medium-th-Unixcape

Whisper Medium TH - Custom datasets and Common voice 17

This model is a fine-tuned version of openai/whisper-medium on the mozilla-foundation/common_voice_17_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1381
• Wer: 59.0037

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: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 3000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2596	0.2406	500	0.2345	75.1845
0.2048	0.4812	1000	0.1912	68.9176
0.164	0.7218	1500	0.1702	66.5990
0.1612	0.9625	2000	0.1496	61.9680
0.075	1.2031	2500	0.1438	59.8647
0.0835	1.4437	3000	0.1381	59.0037

Framework versions

• Transformers 4.45.2
• Pytorch 2.5.1+cu121
• Datasets 3.1.0
• Tokenizers 0.20.3