wav2vec2-large-TIMIT-IPA

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

• Loss: 0.3130
• Per: 0.0550

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: 64
• eval_batch_size: 16
• 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: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Per
4.3003	6.85	500	3.8093	0.9424
1.7151	13.7	1000	0.2929	0.0708
0.2212	20.55	1500	0.2259	0.0575
0.1241	27.4	2000	0.2716	0.0595
0.0917	34.25	2500	0.2902	0.0606
0.0659	41.1	3000	0.2982	0.0570
0.0532	47.95	3500	0.2770	0.0595
0.0438	54.79	4000	0.2953	0.0579
0.0368	61.64	4500	0.3151	0.0572
0.0303	68.49	5000	0.3425	0.0576
0.0281	75.34	5500	0.3065	0.0558
0.0215	82.19	6000	0.3288	0.0558
0.0185	89.04	6500	0.3288	0.0558
0.018	95.89	7000	0.3130	0.0550

Framework versions

• Transformers 4.20.0
• Pytorch 1.12.1+cu113
• Datasets 2.6.2.dev0
• Tokenizers 0.12.1