leenag/Malasar_Medium

This model is a fine-tuned version of openai/whisper-medium on the Spoken Bible Corpus: Malasar dataset. It achieves the following results on the evaluation set:

Loss: 0.5907
Wer: 50.2570

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: 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: 2000
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0646	11.3636	250	0.3369	55.6254
0.0104	22.7273	500	0.4445	52.3130
0.0012	34.0909	750	0.4890	50.1428
0.0002	45.4545	1000	0.5240	50.3712
0.0002	56.8182	1250	0.5488	50.1999
0.0001	68.1818	1500	0.5695	50.3712
0.0001	79.5455	1750	0.5844	50.1999
0.0001	90.9091	2000	0.5907	50.2570

Framework versions

Transformers 4.41.2
Pytorch 2.0.1+cu117
Datasets 2.16.0
Tokenizers 0.19.1

Citation

@misc{multistage2024,
      title={Multistage Fine-tuning Strategies for Automatic Speech Recognition in Low-resource Languages}, 
      author={Leena G Pillai, Kavya Manohar, Basil K Raju, Elizabeth Sherly},
      year={2024},
      eprint={2411.04573},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2411.04573}, 
}