bezaisingh/whisper-medium-bn-cv17

Whisper Medium Bengali

This model is a fine-tuned version of openai/whisper-medium on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0895
• Wer: 82.0349

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: constant_with_warmup
• lr_scheduler_warmup_steps: 500
• training_steps: 6000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1543	0.2618	500	0.1669	48.6957
0.1152	0.5236	1000	0.1206	37.8234
0.0954	0.7853	1500	0.1000	35.2442
0.0535	1.0471	2000	0.0935	36.1322
0.055	1.3089	2500	0.0878	43.8732
0.0537	1.5707	3000	0.0836	55.5653
0.0564	1.8325	3500	0.0776	74.2668
0.0303	2.0942	4000	0.0819	84.9794
0.0291	2.3560	4500	0.0806	56.1679
0.0331	2.6178	5000	0.0784	85.8849
0.0325	2.8796	5500	0.0802	72.0834
0.0158	3.1414	6000	0.0895	82.0349

Framework versions

• Transformers 4.45.2
• Pytorch 2.2.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.1