README.md

metadata

license: apache-2.0
pipeline_tag: automatic-speech-recognition
language:
  - lus
base_model: facebook/wav2vec2-base
tags:
  - mizo
  - audio
  - automatic-speech-recognition
  - lus
datasets:
  - andrewbawitlung/MiZonal-v1.0
metrics:
  - wer
model-index:
  - name: wav2vec2-base-mizo-lus-v25
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: MiZonal v1.0
          type: andrewbawitlung/MiZonal-v1.0
          config: default
          split: train
          args: default
        metrics:
          - name: Wer
            type: wer
            value: 0.16593464935566388

Mizo Automatic Speech Recognition

This model is a fine-tuned version of facebook/wav2vec2-base on the MiZonal v1.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1614
• Wer: 0.1659

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},
issue_date = {July 2025},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
volume = {24},
number = {7},
issn = {2375-4699},
url = {https://doi.org/10.1145/3746063},
doi = {10.1145/3746063},
journal = {ACM Trans. Asian Low-Resour. Lang. Inf. Process.},
month = jul,
articleno = {72},
numpages = {15},
}

Training and evaluation data

MiZonal v1.0.0

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	2.9278	1.0
3.6344	1.45	200	2.8377	1.0
3.6344	2.17	300	2.0445	0.9965
2.1941	2.9	400	0.9115	0.7317
2.1941	3.62	500	0.6427	0.5815
1.0173	4.35	600	0.5384	0.5008
1.0173	5.07	700	0.4707	0.4641
0.7632	5.8	800	0.3804	0.4103
0.7632	6.52	900	0.3635	0.3750
0.6463	7.25	1000	0.3351	0.3670
0.6463	7.97	1100	0.2953	0.3336
0.5674	8.7	1200	0.2711	0.3065
0.5674	9.42	1300	0.2527	0.2877
0.4916	10.14	1400	0.2403	0.2823
0.4916	10.87	1500	0.2352	0.2717
0.442	11.59	1600	0.2312	0.2639
0.442	12.32	1700	0.2251	0.2517
0.4056	13.04	1800	0.1932	0.2275
0.4056	13.77	1900	0.2013	0.2294
0.3726	14.49	2000	0.1954	0.2226
0.3726	15.22	2100	0.1957	0.2175
0.3426	15.94	2200	0.2045	0.2107
0.3426	16.67	2300	0.2003	0.2127
0.3275	17.39	2400	0.1933	0.2023
0.3275	18.12	2500	0.1859	0.2006
0.3112	18.84	2600	0.1821	0.1909
0.3112	19.57	2700	0.1756	0.1888
0.293	20.29	2800	0.1761	0.1865
0.293	21.01	2900	0.1748	0.1990
0.2684	21.74	3000	0.1694	0.1788
0.2684	22.46	3100	0.1745	0.1778
0.2502	23.19	3200	0.1726	0.1739
0.2502	23.91	3300	0.1699	0.1708
0.2435	24.64	3400	0.1670	0.1701
0.2435	25.36	3500	0.1616	0.1714
0.2337	26.09	3600	0.1609	0.1689
0.2337	26.81	3700	0.1615	0.1680
0.2266	27.54	3800	0.1614	0.1659

Framework versions

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