Rishabh06/whisper-small-hi

Whisper Small Hi - Rishabh Mathur

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

• Loss: 0.3956
• WER: 35.9081

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: 16
• total_train_batch_size: 256
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 390
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7731	0.9708	27	0.5895	60.3386
0.4964	1.9775	55	0.4101	45.7155
0.2613	2.9843	83	0.3411	40.6360
0.2032	3.9910	111	0.3155	37.3949
0.1622	4.9978	139	0.3081	36.0648
0.1001	5.9685	166	0.3126	35.4418
0.0826	6.9753	194	0.3265	35.4762
0.0541	7.9820	222	0.3401	35.3348
0.0418	8.9888	250	0.3528	35.3921
0.035	9.9955	278	0.3668	35.4380
0.0245	10.9663	305	0.3783	35.6291
0.0212	11.9730	333	0.3880	36.0304
0.0172	12.9798	361	0.3942	35.8240
0.0159	13.9865	389	0.3956	35.9158
0.0159	14.0225	390	0.3956	35.9081

Framework versions

• Transformers 4.43.4
• Pytorch 2.1.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1