bezaisingh/whisper-base-bn-cv17

Whisper Base Bengali

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

• Loss: 0.1452
• Wer: 43.3377

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
0.3611	0.2618	500	0.3751	80.9274
0.2488	0.5236	1000	0.2592	67.2633
0.2108	0.7853	1500	0.2193	60.3502
0.159	1.0471	2000	0.1998	56.0622
0.1493	1.3089	2500	0.1855	54.1043
0.1455	1.5707	3000	0.1722	51.0905
0.1359	1.8325	3500	0.1636	48.9107
0.1056	2.0942	4000	0.1559	46.7719
0.1035	2.3560	4500	0.1536	45.9615
0.1011	2.6178	5000	0.1506	45.5516
0.0994	2.8796	5500	0.1421	43.5045
0.072	3.1414	6000	0.1452	43.3377

Framework versions

• Transformers 4.45.0
• Pytorch 2.2.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0