kattojuprashanth238/whisper-small-te-v9

Whisper Small Te - Prashanth Kattoju

This model is a fine-tuned version of kattojuprashanth238/whisper-small-te-v6 on the indic-voice dataset. It achieves the following results on the evaluation set:

• Loss: 0.2445
• Wer Ortho: 50.9174
• Wer: 18.3920

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: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• 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: constant_with_warmup
• lr_scheduler_warmup_steps: 50
• training_steps: 1500

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.1999	0.3788	100	0.2231	62.9969	20.0210
0.1478	0.7576	200	0.1937	53.3639	18.5497
0.0719	1.1364	300	0.2053	52.2936	18.9175
0.0851	1.5152	400	0.1674	51.0703	16.7630
0.0682	1.8939	500	0.1752	53.5168	17.2359
0.035	2.2727	600	0.1967	50.0	20.0210
0.0348	2.6515	700	0.2017	53.5168	17.9191
0.0298	3.0303	800	0.2034	51.0703	17.4461
0.0202	3.4091	900	0.2225	55.1988	17.9716
0.02	3.7879	1000	0.2486	56.8807	21.1771
0.0133	4.1667	1100	0.2436	55.6575	23.3841
0.0136	4.5455	1200	0.2337	54.1284	19.6532
0.0116	4.9242	1300	0.2502	56.2691	21.0194
0.0117	5.3030	1400	0.2338	50.0	18.1293
0.0158	5.6818	1500	0.2445	50.9174	18.3920

Framework versions

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