jonatasgrosman/whisper-small-pt-cv11-v6

Whisper Small Portuguese

This model is a fine-tuned version of openai/whisper-small on the mozilla-foundation/common_voice_11_0 pt dataset. It achieves the following results on the evaluation set:

• Loss: 0.2738
• Wer: 11.9723
• Cer: 4.8273

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: 5e-06
• train_batch_size: 32
• 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: 1000
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.5235	0.92	500	0.3605	15.4083	6.1205
0.3839	1.84	1000	0.3034	14.2835	5.6010
0.2828	2.76	1500	0.2852	13.4977	5.1727
0.2367	3.68	2000	0.2768	12.9122	5.2280
0.1832	4.6	2500	0.2728	12.2496	4.9157
0.1549	5.52	3000	0.2730	12.0647	4.8384
0.1318	6.45	3500	0.2757	12.0955	4.8135
0.1077	7.37	4000	0.2738	11.9723	4.8273
0.0969	8.29	4500	0.2784	12.1572	4.9212
0.0813	9.21	5000	0.2805	12.3112	5.0207
0.0751	10.13	5500	0.2831	12.0801	4.8494

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.12.1+cu116
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2