wav2vec2_xls_r_300m_NCHLT_Speech_corpus_Afrikaans_1hr_v4

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: 0.4122
• Wer: 0.9145
• Cer: 0.1750

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: 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	7.5634	1.0	1.0
12.2677	2.0	147	3.7389	1.0	1.0
3.8421	2.9932	220	3.1582	1.0	1.0
3.8421	4.0	294	2.9947	1.0	1.0
3.09	4.9932	367	2.9532	1.0	1.0
2.9584	6.0	441	2.8278	1.0	1.0
2.5616	6.9932	514	1.3050	1.0	0.4732
2.5616	8.0	588	0.8087	0.9809	0.2292
1.1237	8.9932	661	0.6546	0.9642	0.1936
0.6956	10.0	735	0.5676	0.9483	0.1843
0.5116	10.9932	808	0.5192	0.9316	0.1682
0.5116	12.0	882	0.4793	0.9324	0.1744
0.3851	12.9932	955	0.4700	0.9452	0.1816
0.3225	14.0	1029	0.4641	0.9269	0.1639
0.2786	14.9932	1102	0.4991	0.9348	0.1708
0.2786	16.0	1176	0.4696	0.9444	0.1655
0.24	16.9932	1249	0.4680	0.9563	0.1843
0.2253	18.0	1323	0.4289	0.9285	0.1581
0.2253	18.9932	1396	0.4622	0.9308	0.1664
0.2098	20.0	1470	0.4440	0.9491	0.1840
0.1898	20.9932	1543	0.4318	0.9356	0.1631
0.1737	22.0	1617	0.4624	0.9690	0.1935
0.1737	22.9932	1690	0.4387	0.9491	0.1802
0.1507	24.0	1764	0.4203	0.9610	0.1868

Framework versions

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