---
license: apache-2.0
language:
- new
datasets:
- ilprl-docse/Nwacha_Muna_A_Newari_ASR_Dataset
metrics:
- cer
- wer
pipeline_tag: automatic-speech-recognition
tags:
- asr
- nepal-bhasha
- newari
- conformer
- low-resource
- decoder-only
---

## Nwāchā Munā NepConformer Decoder-Only Approach

This model was trained as part of the paper [Nwāchā Munā: A Devanagari Speech Corpus and Proximal Transfer Benchmark for Nepal Bhasha ASR](https://arxiv.org/abs/2603.07554). It is a [NepConformer](https://link.springer.com/chapter/10.1007/978-981-95-2872-1_13) fine-tuned on the [Nwāchā Munā](https://huggingface.co/datasets/ilprl-docse/Nwacha_Muna_A_Newari_ASR_Dataset) corpus with a **decoder-only approach** — achieving **18.77% CER**. The training scripts can be found at [github.com/ilprl/nwacha-muna](https://github.com/ilprl/nwacha-muna).

## Usage
```python
import nemo.collections.asr as nemo_asr

model = nemo_asr.models.ASRModel.from_pretrained("ilprl-docse/NwachaMuna-NepConformer-DO")
transcriptions = model.transcribe(["file.wav"])
print(transcriptions[0])
```