Hungarians/whisper-small-cv16-hu

Whisper Small Hungarian (training in progress)

This model is a fine-tuned version of openai/whisper-small on the Common Voice 16 dataset of Mozilla Foundation. It achieves the following results on the evaluation set:

Tempolary at step 3500:

• Wer: 18.8314

Unfortunatly the colab disconected, this is the end... :( maybe later continue

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: 1.25e-05
• train_batch_size: 8
• eval_batch_size: 4
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 400
• planed training_steps: 6000
• executed steps: 3500 only (colab dc)
• mixed_precision_training: Native AMP

Training results

Steps	Training Loss	Validation Loss	Wer Ortho	Wer
500	0.354600	0.349688	34.385555	31.246555
1000	0.283800	0.290485	29.696507	26.625776
1500	0.248800	0.255122	26.360826	23.300925
2000	0.198300	0.234539	24.557530	21.714145
2500	0.196300	0.224310	23.557423	20.698512
3000	0.153000	0.210894	22.088291	19.231356
3500	0.109100	0.210817	21.465313	18.831435

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.0
• Tokenizers 0.15.0