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

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: 0.2712
• Wer: 17.3335
• Wer Normalized: 11.3321

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
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Wer Normalized
0.2591	0.37	500	0.2776	20.0727	13.6707
0.2626	0.74	1000	0.2599	19.4005	13.3337
0.1131	1.11	1500	0.2516	18.0414	12.1330
0.1016	1.48	2000	0.2482	18.3597	11.9244
0.1094	1.85	2500	0.2411	17.4192	11.6017
0.0524	2.22	3000	0.2512	17.3546	11.4637
0.0433	2.59	3500	0.2496	17.0895	11.2984
0.0453	2.96	4000	0.2479	17.0362	11.2679
0.0201	3.33	4500	0.2693	17.6632	11.7109
0.0229	3.7	5000	0.2712	17.3335	11.3321

Framework versions

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