wav2vec2-base-100h-timit-demo-google-colab

This model is a fine-tuned version of facebook/wav2vec2-base-100h on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2661
• Wer: 0.3003

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: 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: 1000
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.002	2.01	500	3.8351	1.0
2.3505	4.02	1000	0.3927	0.4109
1.3738	6.02	1500	0.3139	0.3470
1.1562	8.03	2000	0.2850	0.3192
1.0548	10.04	2500	0.2677	0.3114
1.0112	12.05	3000	0.2758	0.3103
0.9481	14.06	3500	0.2806	0.3074
0.9109	16.06	4000	0.3118	0.3085
0.8542	18.07	4500	0.2772	0.3040
0.8117	20.08	5000	0.2650	0.3020
0.8058	22.09	5500	0.2848	0.3018
0.7619	24.1	6000	0.2686	0.3018
0.7529	26.1	6500	0.2713	0.3008
0.7172	28.11	7000	0.2661	0.3003

Framework versions

• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 1.18.3
• Tokenizers 0.15.2