Panga-Azazia/dogon-asr

wav2vec2-malian-languages-dogon-dataset

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the SUDOPING01/MALIAN-LANGUAGES-DATASET - DOGON dataset. It achieves the following results on the evaluation set:

• Loss: 0.1688
• Wer: 0.1320

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: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 15.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.4890	100	3.3308	1.0
No log	0.9780	200	2.6855	1.0
No log	1.4645	300	0.4930	0.4569
No log	1.9535	400	0.2583	0.2946
2.4539	2.4401	500	0.2015	0.2439
2.4539	2.9291	600	0.1660	0.1912
2.4539	3.4156	700	0.1410	0.1676
2.4539	3.9046	800	0.1337	0.1610
2.4539	4.3912	900	0.1225	0.1531
0.1717	4.8802	1000	0.1195	0.1402
0.1717	5.3667	1100	0.1282	0.1479
0.1717	5.8557	1200	0.1198	0.1393
0.1717	6.3423	1300	0.1141	0.1334
0.1717	6.8313	1400	0.1167	0.1285
0.1059	7.3178	1500	0.1103	0.1344
0.1059	7.8068	1600	0.1121	0.1230
0.1059	8.2934	1700	0.1206	0.1343
0.1059	8.7824	1800	0.1110	0.1242
0.1059	9.2689	1900	0.1125	0.1238
0.073	9.7579	2000	0.1196	0.1275
0.073	10.2445	2100	0.1223	0.1204
0.073	10.7335	2200	0.1195	0.1200
0.073	11.2200	2300	0.1223	0.1225
0.073	11.7090	2400	0.1164	0.1204
0.0574	12.1956	2500	0.1211	0.1203
0.0574	12.6846	2600	0.1259	0.1213
0.0574	13.1711	2700	0.1312	0.1285
0.0574	13.6601	2800	0.1399	0.1210
0.0574	14.1467	2900	0.1486	0.1272
0.0852	14.6357	3000	0.1698	0.1280

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu121
• Datasets 3.3.1
• Tokenizers 0.21.0