240626-wav2vec2-ASR_Global

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

• Loss: 0.4971
• Wer: 0.0966

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: 5e-05
• train_batch_size: 5
• 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: 500
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	1.4085	100	0.5435	0.1003
No log	2.8169	200	0.5230	0.1092
No log	4.2254	300	0.5379	0.1105
No log	5.6338	400	0.5906	0.1244
0.1133	7.0423	500	0.5424	0.1148
0.1133	8.4507	600	0.5311	0.1318
0.1133	9.8592	700	0.5263	0.12
0.1133	11.2676	800	0.5259	0.1123
0.1133	12.6761	900	0.5031	0.1030
0.128	14.0845	1000	0.5482	0.1103
0.128	15.4930	1100	0.5225	0.1038
0.128	16.9014	1200	0.4823	0.0980
0.128	18.3099	1300	0.4971	0.0966
0.128	19.7183	1400	0.5219	0.0982

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.1+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1