whisper-tiny-ft-cy-en

This model is a fine-tune of openai/whisper-tiny using custom splits from Common Voice 16.1 Welsh and English datasets as well as normalized verbatim transcriptions from techiaith/banc-trawsgrifiadau-bangor

Intended uses & limitations

Due to its small size, this model is intended to be used as the basis for offline speech recognition on devices such as Android phones.

Training and evaluation data

It achieves the following results on the evaluation set:

• Loss: 0.7176
• Wer: 53.1135

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• 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: 4000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.8115	1.41	1000	0.8426	60.0795
0.6396	2.83	2000	0.7508	54.4259
0.5259	4.24	3000	0.7255	53.1328
0.4854	5.66	4000	0.7176	53.1135

Framework versions

• Transformers 4.37.2
• Pytorch 2.2.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1