wav2vec2-base-timit-demo-colab

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

• Loss: 0.5127
• Wer: 0.3082

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
3.7645	2.01	500	2.5179	0.9999
1.1873	4.02	1000	0.5464	0.4798
0.46	6.02	1500	0.4625	0.4025
0.2869	8.03	2000	0.4252	0.3650
0.2213	10.04	2500	0.4340	0.3585
0.1905	12.05	3000	0.4310	0.3404
0.1545	14.06	3500	0.4547	0.3381
0.1206	16.06	4000	0.4902	0.3384
0.1116	18.07	4500	0.4767	0.3253
0.0925	20.08	5000	0.5248	0.3160
0.0897	22.09	5500	0.4960	0.3126
0.0687	24.1	6000	0.4876	0.3086
0.063	26.1	6500	0.4895	0.3065
0.0558	28.11	7000	0.5127	0.3082

Framework versions

• Transformers 4.11.3
• Pytorch 1.10.0+cu111
• Datasets 1.16.1
• Tokenizers 0.10.3