wav2vec2-base-libir-zenodo

This model is a fine-tuned version of facebook/wav2vec2-base-960h on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.4238
• Wer: 0.4336

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: 1e-05
• train_batch_size: 2
• 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
3.053	1.0	31	3.1494	0.7345
2.9742	2.0	62	3.0527	0.7257
2.9139	3.0	93	2.8808	0.7257
2.6586	4.0	124	2.6648	0.6726
2.7117	5.0	155	2.4695	0.6372
2.5173	6.0	186	2.3087	0.6195
2.3665	7.0	217	2.2745	0.6018
2.1276	8.0	248	2.2180	0.5752
2.1624	9.0	279	2.1311	0.5752
2.0312	10.0	310	2.0358	0.5575
2.0652	11.0	341	1.9146	0.5310
1.7963	12.0	372	1.8346	0.5221
1.6811	13.0	403	1.8351	0.5398
1.5929	14.0	434	1.8256	0.4779
1.6644	15.0	465	1.7572	0.4779
1.5411	16.0	496	1.8740	0.4779
1.4027	17.0	527	1.5143	0.4779
1.2634	18.0	558	1.3864	0.4867
1.1053	19.0	589	1.3192	0.4425
1.0517	20.0	620	1.4705	0.4602
1.1033	21.0	651	1.6006	0.4956
0.9992	22.0	682	1.4748	0.5044
0.8987	23.0	713	1.3544	0.4867
0.9656	24.0	744	1.2673	0.4336
0.952	25.0	775	1.3955	0.4071
0.8507	26.0	806	1.3520	0.4425
0.8269	27.0	837	1.8992	0.4336
0.7255	28.0	868	1.9850	0.4425
0.8269	29.0	899	3.0089	0.4425
0.6178	30.0	930	1.4238	0.4336

Framework versions

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