---
language:
- lus
license: apache-2.0
pipeline_tag: automatic-speech-recognition
base_model: Qwen/Qwen3-ASR-1.7B
tags:
- generated_from_trainer
datasets:
- andrewbawitlung/MiZonal-v3.0
metrics:
- wer
- cer
model-index:
- name: qwen3-asr-1.7b-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: 22.6171
    - name: Cer
      type: cer
      value: 4.9795
    - name: Real Time Factor
      type: rtf
      value: 0.0616
---
![Mizo Automatic Speech Recognition (ASR) Models v3.0](banner.jpg)

# qwen3-asr-1.7b-mizonal3-E2-lus-v2026.06

This model is a fine-tuned version of [Qwen/Qwen3-ASR-1.7B](https://huggingface.co/Qwen/Qwen3-ASR-1.7B) on the **MiZonal v3.0** dataset.

It achieves the following results on the evaluation set:
- Wer: 22.6171
- Cer: 4.9795
- Real Time Factor: 0.0616

## Quick Inference


```python
import torch
from qwen_asr import Qwen3ASRModel

# Load the model
model = Qwen3ASRModel.from_pretrained(
    "andrewbawitlung/qwen3-asr-1.7b-mizonal3-E2-lus-v2026.06", 
    dtype=torch.bfloat16, 
    device_map="cuda:0"  # Adjust device as needed
)

# Transcribe audio
results = model.transcribe("your_audio.wav")
print(results)
```


## 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-1.7b-mizonal3-E1-lus-v2026.06](https://huggingface.co/andrewbawitlung/qwen3-asr-1.7b-mizonal3-E1-lus-v2026.06) |
| **E2 (Noise)** | [andrewbawitlung/qwen3-asr-1.7b-mizonal3-E2-lus-v2026.06](https://huggingface.co/andrewbawitlung/qwen3-asr-1.7b-mizonal3-E2-lus-v2026.06) |
| **E3 (Speed)** | [andrewbawitlung/qwen3-asr-1.7b-mizonal3-E3-lus-v2026.06](https://huggingface.co/andrewbawitlung/qwen3-asr-1.7b-mizonal3-E3-lus-v2026.06) |
| **E4 (SpecAug)** | [andrewbawitlung/qwen3-asr-1.7b-mizonal3-E4-lus-v2026.06](https://huggingface.co/andrewbawitlung/qwen3-asr-1.7b-mizonal3-E4-lus-v2026.06) |
| **E5 (Combined)** | [andrewbawitlung/qwen3-asr-1.7b-mizonal3-E5-lus-v2026.06](https://huggingface.co/andrewbawitlung/qwen3-asr-1.7b-mizonal3-E5-lus-v2026.06) |

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 32
- 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 | 1.4517 | 0.4690 | 0.2704 | 23.2461 | 5.2357 | 1.67e-05 | 6.9688 |
| 400 | 2.9035 | 0.2158 | 0.2535 | 20.4048 | 4.5148 | 1.30e-05 | 6.1562 |
| 600 | 4.3497 | 0.0796 | 0.2904 | 20.5799 | 4.8010 | 9.34e-06 | 4.6875 |
| 800 | 5.8015 | 0.0518 | 0.3153 | 21.3876 | 5.1986 | 5.64e-06 | 3.4844 |
| 1000 | 7.2477 | 0.0325 | 0.3302 | 21.2708 | 4.9724 | 1.94e-06 | 2.4062 |