Skier8402/whisper-small-tiny

Whisper Tiny Sw - Skier8402

This model is a fine-tuned version of openai/whisper-tiny on the Common Voice 13 dataset using the swahili only.

Model description

More information needed.

Intended uses & limitations

The model was trained without enough noise added as a form of data augmentation. Do not use this production. I recommend using a larger version of whisper with more hyperparameter tuning especially the learning rate, momentum, weight decay and adjusting the batch size.

Training and evaluation data

I followed the tutorial here. Very minimum edits to the code were done following this tutorial.

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
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 50
• training_steps: 500

Framework versions

• Transformers 4.34.1
• Pytorch 2.1.0+cu121
• Datasets 2.14.5
• Tokenizers 0.14.1