README.md

metadata

license: apache-2.0
language:
  - lus
base_model: facebook/wav2vec2-xls-r-300m
tags:
  - mizo
  - audio
  - automatic-speech-recognition
  - lus
metrics:
  - wer
model-index:
  - name: wav2vec2-xls-r-300m-mizo-lus-v13
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: generator
          type: generator
          config: default
          split: train
          args: default
        metrics:
          - name: Wer
            type: wer
            value: 0.11839374487185675

Mizo Automatic Speech Recognition

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MiZonal v1.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0932
• Wer: 0.1184

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

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 8
• eval_batch_size: 8
• seed: 49
• gradient_accumulation_steps: 8
• 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	0.73	100	3.2655	1.0
4.2561	1.45	200	2.8818	1.0
4.2561	2.18	300	2.8428	1.0
2.8118	2.9	400	2.3670	0.9994
2.8118	3.63	500	0.8009	0.7144
1.4174	4.35	600	0.4873	0.5069
1.4174	5.08	700	0.3496	0.4169
0.754	5.8	800	0.2846	0.3422
0.754	6.53	900	0.2319	0.3116
0.5884	7.25	1000	0.2122	0.2833
0.5884	7.98	1100	0.1931	0.2655
0.4894	8.7	1200	0.1651	0.2221
0.4894	9.43	1300	0.1520	0.2100
0.4171	10.15	1400	0.1379	0.1925
0.4171	10.88	1500	0.1271	0.1793
0.3695	11.6	1600	0.1199	0.1763
0.3695	12.33	1700	0.1217	0.1712
0.3415	13.06	1800	0.1158	0.1640
0.3415	13.78	1900	0.1142	0.1605
0.3094	14.51	2000	0.1137	0.1530
0.3094	15.23	2100	0.1084	0.1454
0.2829	15.96	2200	0.1045	0.1464
0.2829	16.68	2300	0.1025	0.1416
0.2641	17.41	2400	0.0998	0.1374
0.2641	18.13	2500	0.0987	0.1461
0.2486	18.86	2600	0.0937	0.1332
0.2486	19.58	2700	0.0972	0.1337
0.2338	20.31	2800	0.0949	0.1322
0.2338	21.03	2900	0.0982	0.1313
0.2143	21.76	3000	0.0958	0.1311
0.2143	22.48	3100	0.0960	0.1252
0.2018	23.21	3200	0.0930	0.1251
0.2018	23.93	3300	0.0924	0.1243
0.1933	24.66	3400	0.0931	0.1225
0.1933	25.39	3500	0.0942	0.1197
0.1813	26.11	3600	0.0938	0.1208
0.1813	26.84	3700	0.0936	0.1199
0.1792	27.56	3800	0.0932	0.1184

Framework versions

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