whisper-base-en-india-accent-svarah

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

• Loss: 0.3400
• Wer: 16.3057

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: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• 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
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.8871	1.0	47	0.8439	26.5605
0.5938	2.0	94	0.4767	21.4809
0.402	3.0	141	0.4090	18.8854
0.3359	4.0	188	0.3824	17.8503
0.2878	5.0	235	0.3632	17.4841
0.2416	6.0	282	0.3505	16.9904
0.1986	7.0	329	0.3422	16.7834
0.1596	8.0	376	0.3400	16.3057
0.1232	9.0	423	0.3427	16.6242
0.0901	10.0	470	0.3610	16.7357

Framework versions

• Transformers 4.46.3
• Pytorch 2.5.1+cu121
• Datasets 3.1.0
• Tokenizers 0.20.3