somu9/whisper-small-alb

Whisper Small Alb - Sumitesh

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: 1.2013
• Wer: 52.6332

Model description

This is a speech to text model finetuned over Whisper model by OpenAI.

Intended uses & limitations

This is free to use for learning or commercial purposes. I don't plan to monetize this ever or make it private. My goal is to make whisper more localized which is why i have this trained this model and made it public for everyone.

Training and evaluation data

This model is trained on common_voice_17 dataset. It is an open source multilingual dataset.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.005	15.1515	1000	0.9955	53.7437
0.0003	30.3030	2000	1.1066	52.5698
0.0001	45.4545	3000	1.1585	52.8553
0.0001	60.6061	4000	1.1889	52.7284
0.0001	75.7576	5000	1.2013	52.6332

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.0+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1