Whisper Small Sr JV

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

Loss: 0.8988
Wer Ortho: 0.4591
Wer: 0.3415

Model description

Dataset Juzne vesti is published by

Rupnik, Peter and Ljubešić, Nikola, 2022,
ASR training dataset for Serbian JuzneVesti-SR v1.0, Slovenian language resource repository CLARIN.SI, ISSN 2820-4042,
http://hdl.handle.net/11356/1679.

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: 4000
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.5956	0.74	500	0.8063	0.4794	0.3674
0.4401	1.48	1000	0.7975	0.4574	0.3423
0.3029	2.22	1500	0.7821	0.4512	0.3392
0.3016	2.96	2000	0.7828	0.4497	0.3318
0.2372	3.7	2500	0.8254	0.4503	0.3335
0.1762	4.44	3000	0.8402	0.4505	0.3381
0.1414	5.18	3500	0.8945	0.4584	0.3418
0.1326	5.92	4000	0.8988	0.4591	0.3415

Framework versions

Transformers 4.35.2
Pytorch 2.0.1+cu117
Datasets 2.14.5
Tokenizers 0.14.1

Sagicc
/

whisper-small-sr-jv