Whisper Small Italian

This model is a fine-tuned version of openai/whisper-small on the Common Voice 11.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2421
• Wer: 17.3709

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: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 2
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.4521	0.1	100	1.3771	120.3480
0.7526	0.21	200	0.9120	83.8949
0.3023	0.31	300	0.4427	26.2063
0.2718	0.42	400	0.4282	25.9013
0.2823	0.52	500	0.4181	26.2757
0.3151	0.63	600	0.4095	25.0624
0.2559	0.73	700	0.4028	25.4784
0.2727	0.84	800	0.2888	19.5491
0.2532	0.94	900	0.2779	19.3832
0.232	1.05	1000	0.2722	18.6778
0.2169	1.15	1100	0.2720	18.9268
0.2493	1.26	1200	0.2741	20.0678
0.2312	1.36	1300	0.2666	18.2767
0.2158	1.47	1400	0.2651	19.6529
0.2171	1.57	1500	0.2583	18.6087
0.2074	1.68	1600	0.2551	17.6820
0.1862	1.78	1700	0.2491	17.4124
0.2044	1.89	1800	0.2475	17.8964
0.1877	1.99	1900	0.2421	17.3709

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu116
• Datasets 2.7.1
• Tokenizers 0.13.2