wav2vec2_xls_r_300m_NCHLT_Speech_corpus_Afrikaans_1hr_v3

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: 0.0114
• Wer: 0.0809
• Cer: 0.0033

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	8.1690	1.0	1.0
12.6957	2.0	147	4.1344	1.0	1.0
4.1664	2.9932	220	3.3054	1.0	1.0
4.1664	4.0	294	3.0289	1.0	1.0
3.1754	4.9932	367	2.9564	1.0	1.0
2.9736	6.0	441	2.8791	1.0	0.9997
2.8044	6.9932	514	1.8803	1.0	0.8299
2.8044	8.0	588	0.7185	0.9804	0.2074
1.3763	8.9932	661	0.4408	0.8604	0.1121
0.7173	10.0	735	0.2930	0.7089	0.0736
0.5056	10.9932	808	0.1949	0.5855	0.0500
0.5056	12.0	882	0.1374	0.4570	0.0318
0.3858	12.9932	955	0.1084	0.4043	0.0246
0.2925	14.0	1029	0.0856	0.3489	0.0202
0.2587	14.9932	1102	0.0738	0.2894	0.0200
0.2587	16.0	1176	0.0524	0.2094	0.0136
0.2164	16.9932	1249	0.0450	0.2119	0.0113
0.1994	18.0	1323	0.0430	0.2094	0.0103
0.1994	18.9932	1396	0.0317	0.1872	0.0100
0.1665	20.0	1470	0.0323	0.1557	0.0077
0.1472	20.9932	1543	0.0248	0.0953	0.0052
0.1303	22.0	1617	0.0268	0.1328	0.0063
0.1303	22.9932	1690	0.0236	0.1957	0.0103
0.117	24.0	1764	0.0197	0.1455	0.0064
0.1066	24.9932	1837	0.0169	0.1362	0.0066
0.0976	26.0	1911	0.0168	0.1983	0.0106
0.0976	26.9932	1984	0.0114	0.0809	0.0033
0.0932	28.0	2058	0.0116	0.1055	0.0051
0.0846	28.9932	2131	0.0098	0.0987	0.0046
0.0799	30.0	2205	0.0099	0.0953	0.0046
0.0799	30.9932	2278	0.0084	0.1864	0.0096
0.0768	32.0	2352	0.0093	0.16	0.0083
0.0709	32.9932	2425	0.0070	0.1574	0.0081
0.0709	34.0	2499	0.0074	0.1311	0.0062
0.0662	34.9932	2572	0.0072	0.1098	0.0051
0.0612	36.0	2646	0.0059	0.1804	0.0091
0.0603	36.9932	2719	0.0049	0.1319	0.0082

Framework versions

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