wav2vec2-vivos

This model is a fine-tuned version of facebook/wav2vec2-base on the vivos dataset. It achieves the following results on the evaluation set:

• Loss: 0.4598
• Wer: 0.2343

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: 0.0002
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.25
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.8271	2.0	146	3.8747	1.0
3.4616	4.0	292	3.5849	1.0
3.35	6.0	438	2.6294	0.9997
1.1993	8.0	584	0.6472	0.4255
0.4734	10.0	730	0.5342	0.3258
0.3156	12.0	876	0.4651	0.2758
0.2392	14.0	1022	0.4690	0.2573
0.2183	16.0	1168	0.4601	0.2434
0.164	18.0	1314	0.4619	0.2379
0.1452	20.0	1460	0.4598	0.2343

Framework versions

• Transformers 4.44.0
• Pytorch 2.4.0
• Datasets 2.21.0
• Tokenizers 0.19.1