Stopwolf/whisper-small-sr

Whisper Small Serbian

This model is a fine-tuned version of openai/whisper-small on the Common Voice 13 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4671
• Wer Ortho: 27.4565
• Wer: 17.4196

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: 2500

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.1403	1.44	250	0.2809	28.8913	19.2224
0.0664	2.87	500	0.2858	27.3696	17.9626
0.0315	4.31	750	0.3152	27.9348	17.4631
0.0174	5.75	1000	0.3578	28.1522	17.9844
0.0067	7.18	1250	0.4018	27.9130	17.9626
0.0015	8.62	1500	0.4535	28.6739	17.5717
0.0008	10.06	1750	0.4558	27.2174	17.1807
0.0005	11.49	2000	0.4585	27.4348	17.4848
0.0005	12.93	2250	0.4651	27.3478	17.3979
0.0005	14.37	2500	0.4671	27.4565	17.4196

Framework versions

• Transformers 4.33.3
• Pytorch 2.0.1+cu118
• Datasets 2.14.5
• Tokenizers 0.13.3