w2v-bert-2.0-clearglobal-kanuri-asr-1.0.0

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the TWB Voice 1.0 dataset.

It achieves the following results on the evaluation set:

• WER: 11.31%
• CER: 3.05%

Training and evaluation data

This model was trained by colleagues from the Makerere University Centre for Artificial Intelligence and Data Science in collaboration with CLEAR Global. We gratefully acknowledge their expertise and partnership.

Model was trained and tested on the approved Kanuri subset of TWB Voice 1.0 dataset.

Train/dev/test portions correspond to the splits in this dataset version.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Use OptimizerNames.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_ratio: 0.08
• num_epochs: 50.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
2.4049	1.0	247	0.4777	0.6218	0.1402
0.4273	2.0	494	0.3793	0.5352	0.1183
0.3533	3.0	741	0.3585	0.4794	0.1077
0.3232	4.0	988	0.3429	0.4771	0.1069
0.2906	5.0	1235	0.3269	0.4533	0.1013
0.2638	6.0	1482	0.3006	0.4144	0.0909
0.232	7.0	1729	0.3044	0.4137	0.0908
0.2094	8.0	1976	0.2940	0.4078	0.0891
0.1803	9.0	2223	0.2819	0.3906	0.0833
0.1573	10.0	2470	0.2761	0.3533	0.0768
0.1307	11.0	2717	0.2710	0.3092	0.0670
0.0967	12.0	2964	0.2467	0.2900	0.0625
0.075	13.0	3211	0.2471	0.2715	0.0578
0.057	14.0	3458	0.2396	0.2449	0.0527
0.0454	15.0	3705	0.2496	0.2365	0.0526
0.0369	16.0	3952	0.2429	0.2134	0.0463
0.027	17.0	4199	0.2331	0.2002	0.0425
0.0221	18.0	4446	0.2473	0.1945	0.0434
0.0212	19.0	4693	0.2542	0.2054	0.0452
0.0203	20.0	4940	0.2456	0.1963	0.0426
0.017	21.0	5187	0.2313	0.1687	0.0374
0.0133	22.0	5434	0.2517	0.1764	0.0397
0.0122	23.0	5681	0.2450	0.1787	0.0402
0.0107	24.0	5928	0.2646	0.1730	0.0392
0.0104	25.0	6175	0.2628	0.1702	0.0378
0.0079	26.0	6422	0.2735	0.1649	0.0371
0.0062	27.0	6669	0.2478	0.1476	0.0339
0.0075	28.0	6916	0.2541	0.1800	0.0408
0.0109	29.0	7163	0.2437	0.1563	0.0354
0.0068	30.0	7410	0.2314	0.1411	0.0318
0.0048	31.0	7657	0.2665	0.1455	0.0332
0.0056	32.0	7904	0.2366	0.1507	0.0345
0.0043	33.0	8151	0.2444	0.1376	0.0317
0.0029	34.0	8398	0.2480	0.1341	0.0306
0.0027	35.0	8645	0.2517	0.1355	0.0310
0.0025	36.0	8892	0.2495	0.1259	0.0293
0.0011	37.0	9139	0.2563	0.1259	0.0295
0.0009	38.0	9386	0.2708	0.1313	0.0307
0.0018	39.0	9633	0.2672	0.1353	0.0317
0.0024	40.0	9880	0.2506	0.1372	0.0322
0.0013	41.0	10127	0.2556	0.1255	0.0291

Framework versions

• Transformers 4.53.1
• Pytorch 2.7.1+cu126
• Datasets 4.0.0
• Tokenizers 0.21.2