whisper-ling-asr

This model is a fine-tuned version of openai/whisper-tiny on the KasuleTrevor/lingala_20hr dataset. It achieves the following results on the evaluation set:

• Loss: 0.4835
• Wer: 0.2979

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
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• training_steps: 12000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3703	2.9851	1000	0.5100	0.2735
0.1487	5.9701	2000	0.4835	0.2979
0.0556	8.9552	3000	0.5050	0.2544
0.0171	11.9403	4000	0.5401	0.2370
0.0051	14.9254	5000	0.5673	0.2177

Framework versions

• Transformers 4.43.0
• Pytorch 2.4.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1