MCDHANUSH/whisper-tiny-ta

Whisper tiny ta - Sanchit Gandhi

This model is a fine-tuned version of openai/whisper-tiny on the Common Voice 11.0 dataset. It achieves the following results on the evaluation set:

• Loss: 1.1323
• Wer: 81.8182
• Cer: 25.8981

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: 0.0001
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.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_steps: 100
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.3684	1.5873	100	0.7662	96.3317	49.9906
0.1868	3.1746	200	0.8380	86.7624	29.6596
0.097	4.7619	300	0.9112	85.0080	29.5091
0.0481	6.3492	400	0.9833	85.4864	29.9041
0.0332	7.9365	500	0.9751	83.0941	30.1862
0.0154	9.5238	600	1.0561	85.4864	29.2082
0.0064	11.1111	700	1.1354	83.5726	27.3462
0.003	12.6984	800	1.1157	83.7321	27.1958
0.0006	14.2857	900	1.1344	82.7751	26.4435
0.0004	15.8730	1000	1.1323	81.8182	25.8981

Framework versions

• Transformers 4.49.0
• Pytorch 2.5.1+cu124
• Datasets 3.3.2
• Tokenizers 0.21.0