nyarkssss/epoch_small_pwer

Whisper Small Akan

This model is a fine-tuned version of openai/whisper-small on the Speech Data Ghana UG - Ghanaian Multilingual Sample Data dataset. It achieves the following results on the evaluation set:

• Loss: 0.9316
• Wer: 40.4973

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: 0.0001
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use adamw_bnb_8bit 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: 200
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3432	2.5	250	0.6637	46.8960
0.1538	5.0	500	0.7127	45.7291
0.0515	7.5	750	0.8340	45.6681
0.0282	10.0	1000	0.8644	42.6480
0.0111	12.5	1250	0.8923	42.6022
0.0016	15.0	1500	0.9055	40.6650
0.0003	17.5	1750	0.9256	40.4667
0.0003	20.0	2000	0.9316	40.4973

Framework versions

• Transformers 4.46.0
• Pytorch 2.4.0
• Datasets 3.0.2
• Tokenizers 0.20.0