andrewbawitlung/wav2vec2-large-mizo-lus-v25

Mizo Automatic Speech Recognition

This model is a fine-tuned version of facebook/wav2vec2-large on the generator dataset. It achieves the following results on the evaluation set: - Loss: 0.1551 - Wer: 0.1937

Citation

BibTeX entry and citation info:

@article{10.1145/3746063,
author = {Bawitlung, Andrew and Dash, Sandeep Kumar and Pattanayak, Radha Mohan},
title = {Mizo Automatic Speech Recognition: Leveraging Wav2vec 2.0 and XLS-R for Enhanced Accuracy in Low-Resource Language Processing},
year = {2025},
url = {https://doi.org/10.1145/3746063},
doi = {10.1145/3746063},
journal = {ACM Trans. Asian Low-Resour. Lang. Inf. Process.},
month = jun,
}

Training and evaluation data

MiZonal v1.0

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 49
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 28
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.72	100	2.9713	1.0
3.3962	1.45	200	2.5742	0.9999
3.3962	2.17	300	1.0420	0.8141
1.5969	2.9	400	0.7051	0.6338
1.5969	3.62	500	0.5743	0.5490
0.9636	4.35	600	0.5414	0.4995
0.9636	5.07	700	0.4623	0.4644
0.804	5.8	800	0.4226	0.4230
0.804	6.52	900	0.4549	0.3944
0.7332	7.25	1000	0.4360	0.3908
0.7332	7.97	1100	0.3846	0.4101
0.666	8.7	1200	0.3979	0.4136
0.666	9.42	1300	0.3188	0.3691
0.6051	10.14	1400	0.2872	0.3473
0.6051	10.87	1500	0.2632	0.3448
0.5467	11.59	1600	0.2454	0.3091
0.5467	12.32	1700	0.2387	0.3390
0.5073	13.04	1800	0.2727	0.2812
0.5073	13.77	1900	0.2328	0.3185
0.4611	14.49	2000	0.2480	0.2777
0.4611	15.22	2100	0.2246	0.2517
0.4237	15.94	2200	0.2243	0.2598
0.4237	16.67	2300	0.2122	0.2719
0.3901	17.39	2400	0.1983	0.2461
0.3901	18.12	2500	0.2150	0.2240
0.3664	18.84	2600	0.2058	0.2310
0.3664	19.57	2700	0.1860	0.2231
0.3352	20.29	2800	0.1691	0.2276
0.3352	21.01	2900	0.1934	0.2206
0.313	21.74	3000	0.1887	0.2182
0.313	22.46	3100	0.1758	0.2060
0.2854	23.19	3200	0.1811	0.2166
0.2854	23.91	3300	0.1667	0.2109
0.2666	24.64	3400	0.1658	0.2005
0.2666	25.36	3500	0.1621	0.1960
0.2604	26.09	3600	0.1599	0.2018
0.2604	26.81	3700	0.1527	0.1968
0.2452	27.54	3800	0.1551	0.1937

Framework versions

• Transformers 4.37.2
• Pytorch 2.3.1+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1