wav2vec2-xls-r-300m-en-phoneme-ctc-60h-noisy

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.1492
• Per: 0.0393
• Phoneme Accuracy: 0.9607

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: 24
• eval_batch_size: 24
• seed: 42
• optimizer: Use adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Per	Phoneme Accuracy
3.8716	0.3674	500	3.7213	1.0	0.0
3.5698	0.7348	1000	3.5655	1.0	0.0
3.2458	1.1021	1500	2.8473	0.9143	0.0857
0.7065	1.4695	2000	0.3996	0.0893	0.9107
0.5249	1.8369	2500	0.2702	0.0657	0.9343
0.4425	2.2043	3000	0.2217	0.0567	0.9433
0.3613	2.5716	3500	0.1995	0.0515	0.9485
0.3521	2.9390	4000	0.1897	0.0485	0.9515
0.3083	3.3064	4500	0.1808	0.0468	0.9532
0.3115	3.6738	5000	0.1823	0.0461	0.9539
0.2905	4.0411	5500	0.1667	0.0448	0.9552
0.2948	4.4085	6000	0.1700	0.0441	0.9559
0.2578	4.7759	6500	0.1682	0.0437	0.9563
0.2351	5.1433	7000	0.1586	0.0424	0.9576
0.244	5.5107	7500	0.1601	0.0422	0.9578
0.2318	5.8780	8000	0.1541	0.0414	0.9586
0.2667	6.2454	8500	0.1557	0.0413	0.9587
0.2381	6.6128	9000	0.1527	0.0405	0.9595
0.2317	6.9802	9500	0.1498	0.0401	0.9599
0.1982	7.3475	10000	0.1518	0.0406	0.9594
0.2075	7.7149	10500	0.1492	0.0404	0.9596
0.2049	8.0823	11000	0.1530	0.0396	0.9604
0.1926	8.4497	11500	0.1501	0.0393	0.9607
0.2129	8.8170	12000	0.1463	0.0395	0.9605
0.2193	9.1844	12500	0.1478	0.0393	0.9607
0.1999	9.5518	13000	0.1483	0.0390	0.9610
0.2192	9.9192	13500	0.1492	0.0393	0.9607

Framework versions

• Transformers 4.57.3
• Pytorch 2.9.1+cu126
• Datasets 4.4.2
• Tokenizers 0.22.1