wav2vec2-base-librispeech

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

• Loss: 0.9548
• Wer: 0.4070

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: 0.0001
• train_batch_size: 4
• 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: 1000
• num_epochs: 150
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.4865	29.41	500	3.5010	1.0
1.112	58.82	1000	1.0382	0.4767
0.111	88.24	1500	0.9833	0.5116
0.0438	117.65	2000	0.9302	0.4302
0.0241	147.06	2500	0.9548	0.4070

Framework versions

• Transformers 4.36.2
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2