thiagobarbosa/whisper-small-common-voice-16-pt-v2

Whisper small using Common Voice 16 (pt)

This model is a fine-tuned version of openai/whisper-small on the Mozilla Common Voices - 16.0 - Portuguese dataset. It achieves the following results on the evaluation set:

• Loss: 0.2220
• Wer: 16.0359
• Wer Normalized: 10.3867

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: 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
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Wer Normalized
0.2484	0.26	500	0.2712	19.2259	13.0929
0.2184	0.52	1000	0.2464	17.8895	11.9404
0.236	0.77	1500	0.2339	17.1348	11.3016
0.1401	1.03	2000	0.2285	16.7001	11.0432
0.1206	1.29	2500	0.2251	16.3235	10.6467
0.1199	1.55	3000	0.2236	16.1732	10.5424
0.1231	1.81	3500	0.2197	16.1587	10.5038
0.0935	2.06	4000	0.2220	16.0359	10.3867

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.2
• Datasets 2.16.1
• Tokenizers 0.15.1