whisper-large-v3-ft-cv-cy-en

This model is a fine-tuned version of openai/whisper-large-v3 on the DewiBrynJones/commonvoice_18_0_cy_en train main dataset. It achieves the following results on the evaluation set:

• Loss: 0.4825
• Wer: 0.1979

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
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• 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
0.2506	2.0060	1000	0.3449	0.2113
0.0783	4.0120	2000	0.3657	0.2010
0.0377	6.0181	3000	0.4253	0.1986
0.0326	8.0241	4000	0.4601	0.1977
0.0313	10.0301	5000	0.4825	0.1979

Framework versions

• Transformers 4.44.0
• Pytorch 2.4.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1