wav2vec2-large-xlsr-53-br

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice_15_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7879
• Wer: 54.7151
• Cer: 19.2493

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: 6e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 300
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
6.3257	2.18	500	3.0700	100.0	99.0871
2.2071	4.36	1000	1.1541	80.0449	29.4230
1.0019	6.54	1500	0.8986	69.2059	24.3938
0.7796	8.71	2000	0.8015	63.3737	22.1296
0.6677	10.89	2500	0.8014	61.4984	21.4568
0.5937	13.07	3000	0.7623	58.9323	20.4929
0.5454	15.25	3500	0.7975	57.8466	20.2585
0.5075	17.43	4000	0.7831	56.7250	19.7879
0.4837	19.61	4500	0.7902	55.9623	19.5101
0.4529	21.79	5000	0.7851	54.9753	19.0924
0.4381	23.97	5500	0.7865	55.1727	19.3211
0.4208	26.14	6000	0.8168	55.1817	19.3967
0.4197	28.32	6500	0.7879	54.7151	19.2493

Framework versions

• Transformers 4.39.1
• Pytorch 2.0.1+cu117
• Datasets 2.18.0
• Tokenizers 0.15.2