wav2vec2_xls_r_300m_NCHLT_Speech_corpus_Afrikaans_1hr_v5

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: 13.2415
• Wer: 1.0
• Cer: 1.0

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
No log	0.9932	73	10.7362	1.0	1.0
14.391	2.0	147	4.2584	1.0	1.0
4.4016	2.9932	220	3.5537	1.0	1.0
4.4016	4.0	294	3.1929	1.0	1.0
3.4024	4.9932	367	3.0606	1.0	1.0
3.0714	6.0	441	3.0551	1.0	1.0
2.967	6.9932	514	2.9778	1.0	1.0
2.967	8.0	588	2.9386	1.0	1.0
2.8302	8.9932	661	2.6021	1.0	0.9055
1.783	10.0	735	1.6877	1.0	0.6387
0.9447	10.9932	808	1.4003	1.0	0.4661

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1