michaelszhu/whisper-small-finetuned-radio-ASR-2

BANG please be the final one (EN)

This model is a fine-tuned version of openai/whisper-large-v3 on the Radio-Modified Common Voice 11.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0395
• Wer: 8.8210

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: 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: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1511	0.25	1000	0.1318	20.4937
0.0685	1.2443	2000	0.0845	12.3199
0.0378	2.2385	3000	0.0557	10.4397
0.0283	3.2328	4000	0.0395	8.8210

Framework versions

• Transformers 4.41.1
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1