asr-africa/wav2vec2_xls_r_300m_DigitalUmuganda_Afrivoice_Shona_86hr_v1

facebook/wav2vec2-xls-r-300m

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

• Loss: 0.3536
• Wer: 0.3289
• Cer: 0.0684

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.8923	1.0	1885	0.3250	0.4671	0.0874
0.2338	2.0	3770	0.2228	0.3238	0.0547
0.1887	3.0	5655	0.2019	0.2795	0.0479
0.1641	4.0	7540	0.1987	0.2836	0.0493
0.1445	5.0	9425	0.1932	0.2673	0.0455
0.1251	6.0	11310	0.1969	0.2710	0.0457
0.1082	7.0	13195	0.2057	0.2623	0.0441
0.0929	8.0	15080	0.2181	0.2676	0.0451
0.0796	9.0	16965	0.2352	0.2678	0.0448
0.0671	10.0	18850	0.2427	0.2725	0.0460
0.0575	11.0	20735	0.2593	0.2709	0.0451
0.0492	12.0	22620	0.2744	0.2789	0.0460
0.0427	13.0	24505	0.2792	0.2763	0.0456
0.0389	14.0	26390	0.2831	0.2871	0.0461
0.034	15.0	28275	0.2987	0.2734	0.0446
0.0322	16.0	30160	0.3052	0.2806	0.0450
0.0298	17.0	32045	0.3148	0.2713	0.0443

Framework versions

• Transformers 4.45.2
• Pytorch 2.2.0+cu121
• Datasets 3.0.1
• Tokenizers 0.20.1