README.md

---
language:
- lus
license: apache-2.0
pipeline_tag: automatic-speech-recognition
base_model: Qwen/Qwen3-ASR-0.6B
tags:
- generated_from_trainer
datasets:
- andrewbawitlung/MiZonal-v3.0
metrics:
- wer
- cer
model-index:
- name: qwen3-asr-0.6b-mizonal3-E2-lus-v2026.06
  results:
  - task:
      name: Automatic Speech Recognition
      type: automatic-speech-recognition
    dataset:
      name: MiZonal v3.0
      type: andrewbawitlung/MiZonal-v3.0
      config: default
      split: test
    metrics:
    - name: Wer
      type: wer
      value: 19.4954
    - name: Cer
      type: cer
      value: 4.4174
    - name: Real Time Factor
      type: rtf
      value: 0.0660
---
![Mizo Automatic Speech Recognition (ASR) Models v3.0](banner.jpg)

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# qwen3-asr-0.6b-mizonal3-E2-lus-v2026.06

This model is a fine-tuned version of [Qwen/Qwen3-ASR-0.6B](https://huggingface.co/Qwen/Qwen3-ASR-0.6B) on the **MiZonal v3.0** dataset. 
Note: ~1 hour of conversational speech was added to this dataset version.

It achieves the following results on the evaluation set:
- Wer: 19.4954
- Cer: 4.4174
- Real Time Factor: 0.0660

## Quick Inference


```python
import torch
import librosa
from transformers import AutoProcessor, Qwen2AudioForConditionalGeneration

device = "cuda" if torch.cuda.is_available() else "cpu"

processor = AutoProcessor.from_pretrained("andrewbawitlung/qwen3-asr-0.6b-mizonal3-E2-lus-v2026.06")
model = Qwen2AudioForConditionalGeneration.from_pretrained("andrewbawitlung/qwen3-asr-0.6b-mizonal3-E2-lus-v2026.06").to(device)

audio, sr = librosa.load("your_audio.wav", sr=16000)

conversation = [
    {"role": "user", "content": [
        {"type": "audio", "audio_url": "your_audio.wav"},
        {"type": "text", "text": "Transcribe the audio:"}
    ]}
]
text = processor.apply_chat_template(conversation, add_generation_prompt=True, tokenize=False)
inputs = processor(text=text, audios=[audio], return_tensors="pt", padding=True)
inputs.input_ids = inputs.input_ids.to(device)

with torch.no_grad():
    generate_ids = model.generate(**inputs, max_length=256)

generate_ids = generate_ids[:, inputs.input_ids.size(1):]
transcription = processor.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
print(transcription)
```


## Model description

### Experiment Configurations
This repository is part of a series of experiments. The different configurations are:
- **E1 (Baseline):** Standard training configuration.
- **E2 (Noise):** Training with background noise augmentation.
- **E3 (Speed):** Training with speed perturbation augmentation.
- **E4 (SpecAug):** Training with SpecAugment (time and frequency masking).
- **E5 (Combined):** Training with a combination of all augmentations.

### All Models in this Family
| Experiment | Hugging Face Repository |
| :--- | :--- |
| **E1 (Baseline)** | [andrewbawitlung/qwen3-asr-0.6b-mizonal3-E1-lus-v2026.06](https://huggingface.co/andrewbawitlung/qwen3-asr-0.6b-mizonal3-E1-lus-v2026.06) |
| **E2 (Noise)** | [andrewbawitlung/qwen3-asr-0.6b-mizonal3-E2-lus-v2026.06](https://huggingface.co/andrewbawitlung/qwen3-asr-0.6b-mizonal3-E2-lus-v2026.06) |
| **E3 (Speed)** | [andrewbawitlung/qwen3-asr-0.6b-mizonal3-E3-lus-v2026.06](https://huggingface.co/andrewbawitlung/qwen3-asr-0.6b-mizonal3-E3-lus-v2026.06) |
| **E4 (SpecAug)** | [andrewbawitlung/qwen3-asr-0.6b-mizonal3-E4-lus-v2026.06](https://huggingface.co/andrewbawitlung/qwen3-asr-0.6b-mizonal3-E4-lus-v2026.06) |
| **E5 (Combined)** | [andrewbawitlung/qwen3-asr-0.6b-mizonal3-E5-lus-v2026.06](https://huggingface.co/andrewbawitlung/qwen3-asr-0.6b-mizonal3-E5-lus-v2026.06) |

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- optimizer: OptimizerNames.ADAMW_TORCH_FUSED
- lr_scheduler_type: SchedulerType.LINEAR
- num_epochs: 8

### Training results

| step | epoch | train_loss | eval_loss | eval_wer | eval_cer | learning_rate | grad_norm |
| --- | --- | --- | --- | --- | --- | --- | --- |
| 200 | 0.18 | 0.4524 | 0.7468 | 52.83 | 16.21 | 1.99e-05 | 14.50 |
| 400 | 0.36 | 0.2727 | 0.4002 | 32.25 | 8.09 | 1.95e-05 | 10.81 |
| 600 | 0.55 | 0.1697 | 0.3365 | 27.96 | 6.43 | 1.90e-05 | 7.56 |
| 800 | 0.73 | 0.2061 | 0.2991 | 24.87 | 5.98 | 1.86e-05 | 5.78 |
| 1000 | 0.91 | 0.1388 | 0.2859 | 24.40 | 5.47 | 1.81e-05 | 7.38 |
| 1200 | 1.09 | 0.0883 | 0.2676 | 22.51 | 5.26 | 1.76e-05 | 5.66 |
| 1400 | 1.28 | 0.0722 | 0.2607 | 21.59 | 5.10 | 1.72e-05 | 3.84 |
| 1600 | 1.46 | 0.0763 | 0.2593 | 21.31 | 4.88 | 1.67e-05 | 4.44 |
| 1800 | 1.64 | 0.0755 | 0.2524 | 20.04 | 4.63 | 1.62e-05 | 8.81 |
| 2000 | 1.82 | 0.0622 | 0.2465 | 20.23 | 4.54 | 1.58e-05 | 4.59 |
| 2200 | 2.00 | 0.0477 | 0.2433 | 19.96 | 4.53 | 1.53e-05 | 3.84 |
| 2400 | 2.19 | 0.0447 | 0.2579 | 20.18 | 4.68 | 1.48e-05 | 5.66 |
| 2600 | 2.37 | 0.0332 | 0.2576 | 19.31 | 4.51 | 1.44e-05 | 3.44 |
| 2800 | 2.55 | 0.0305 | 0.2666 | 19.38 | 4.45 | 1.39e-05 | 7.22 |
| 3000 | 2.73 | 0.0297 | 0.2642 | 19.75 | 4.72 | 1.34e-05 | 4.22 |
| 3200 | 2.91 | 0.0472 | 0.2632 | 18.96 | 4.34 | 1.30e-05 | 4.34 |
| 3400 | 3.10 | 0.0152 | 0.2804 | 19.31 | 4.39 | 1.25e-05 | 4.47 |
| 3600 | 3.28 | 0.0114 | 0.2898 | 19.29 | 4.42 | 1.20e-05 | 2.52 |
| 3800 | 3.46 | 0.0159 | 0.2936 | 18.77 | 4.38 | 1.16e-05 | 2.94 |
| 4000 | 3.64 | 0.0090 | 0.2901 | 19.34 | 4.41 | 1.11e-05 | 2.06 |
| 4200 | 3.83 | 0.0115 | 0.2965 | 19.11 | 4.48 | 1.07e-05 | 4.19 |
| 4400 | 4.01 | 0.0062 | 0.3002 | 18.83 | 4.34 | 1.02e-05 | 5.16 |
| 4600 | 4.19 | 0.0054 | 0.3130 | 18.60 | 4.29 | 9.72e-06 | 1.23 |
| 4800 | 4.37 | 0.0047 | 0.3150 | 19.05 | 4.42 | 9.26e-06 | 1.21 |
| 5000 | 4.55 | 0.0094 | 0.3163 | 19.28 | 4.46 | 8.79e-06 | 0.94 |
| 5200 | 4.74 | 0.0037 | 0.3210 | 19.01 | 4.43 | 8.33e-06 | 1.05 |
| 5400 | 4.92 | 0.0041 | 0.3183 | 19.04 | 4.38 | 7.86e-06 | 0.99 |
| 5600 | 5.10 | 0.0022 | 0.3311 | 19.35 | 4.44 | 7.40e-06 | 0.46 |
| 5800 | 5.28 | 0.0022 | 0.3335 | 19.17 | 4.44 | 6.94e-06 | 0.70 |
| 6000 | 5.46 | 0.0023 | 0.3346 | 18.99 | 4.37 | 6.47e-06 | 0.84 |
| 6200 | 5.65 | 0.0027 | 0.3361 | 19.28 | 4.42 | 6.01e-06 | 0.60 |
| 6400 | 5.83 | 0.0025 | 0.3363 | 19.28 | 4.49 | 5.54e-06 | 2.16 |
| 6600 | 6.01 | 0.0019 | 0.3365 | 19.04 | 4.42 | 5.08e-06 | 0.46 |
| 6800 | 6.19 | 0.0022 | 0.3407 | 19.27 | 4.42 | 4.61e-06 | 0.35 |
| 7000 | 6.38 | 0.0018 | 0.3427 | 19.35 | 4.47 | 4.15e-06 | 0.38 |
| 7200 | 6.56 | 0.0018 | 0.3420 | 19.17 | 4.45 | 3.68e-06 | 0.61 |
| 7400 | 6.74 | 0.0034 | 0.3414 | 19.31 | 4.45 | 3.22e-06 | 0.41 |
| 7600 | 6.92 | 0.0019 | 0.3420 | 19.15 | 4.38 | 2.75e-06 | 0.35 |
| 7800 | 7.10 | 0.0017 | 0.3427 | 19.27 | 4.42 | 2.29e-06 | 0.43 |
| 8000 | 7.29 | 0.0024 | 0.3426 | 19.46 | 4.49 | 1.82e-06 | 0.52 |
| 8200 | 7.47 | 0.0016 | 0.3432 | 19.31 | 4.49 | 1.36e-06 | 0.28 |
| 8400 | 7.65 | 0.0017 | 0.3435 | 19.24 | 4.43 | 8.95e-07 | 0.33 |
| 8600 | 7.83 | 0.0021 | 0.3429 | 19.36 | 4.45 | 4.30e-07 | 0.74 |