wav2vec2-base-2

This model is a fine-tuned version of jiobiala24/wav2vec2-base-1 on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 0.9415
• Wer: 0.3076

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: 32
• 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
0.4206	1.96	1000	0.6022	0.3435
0.3278	3.93	2000	0.6191	0.3344
0.2604	5.89	3000	0.6170	0.3288
0.2135	7.86	4000	0.6590	0.3239
0.1805	9.82	5000	0.7359	0.3289
0.1582	11.79	6000	0.7450	0.3276
0.1399	13.75	7000	0.7914	0.3218
0.1252	15.72	8000	0.8254	0.3185
0.1095	17.68	9000	0.8524	0.3184
0.1	19.65	10000	0.8340	0.3165
0.0905	21.61	11000	0.8846	0.3161
0.0819	23.58	12000	0.8994	0.3142
0.0763	25.54	13000	0.9018	0.3134
0.0726	27.5	14000	0.9552	0.3081
0.0668	29.47	15000	0.9415	0.3076

Framework versions

• Transformers 4.11.3
• Pytorch 1.10.0+cu111
• Datasets 1.13.3
• Tokenizers 0.10.3