kavyamanohar/whisper-supreme-court-asr

This model is a fine-tuned version of openai/whisper-small on the Supreme Court Hearing Corpus - Subset dataset. It achieves the following results on the evaluation set:

• Loss: 1.8703
• Wer: 69.8671

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
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use 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: 500
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.8763	5.5714	100	1.8904	162.7770
0.7801	11.1143	200	1.5403	55.0222
0.0716	16.6857	300	1.5505	56.7947
0.0118	22.2286	400	1.6573	56.8685
0.0076	27.8	500	1.7218	83.3087
0.0014	33.3429	600	1.7909	79.2467
0.0005	38.9143	700	1.8290	67.8730
0.0003	44.4571	800	1.8555	69.4239
0.0002	50.0	900	1.8669	69.7932
0.0002	55.5714	1000	1.8703	69.8671

Framework versions

• Transformers 4.48.2
• Pytorch 2.6.0+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0