Nw\=ach\=a Mun\=a: A Devanagari Speech Corpus and Proximal Transfer Benchmark for Nepal Bhasha ASR

TL;DR

This paper introduces Nw ext=ach ext= a Mun ext= a, a new Devanagari speech corpus for Nepal Bhasha, and demonstrates that proximal transfer from Nepali can effectively enable low-resource ASR.

Contribution

The creation of a 5.39-hour manually transcribed Nepal Bhasha speech corpus and a benchmark for script-preserving acoustic modeling.

Findings

Fine-tuning Nepali Conformer reduces CER from 52.54% to 17.59%.

Proximal transfer from Nepali rivals large multilingual models.

Dataset and benchmarks are openly released.

Abstract

Nepal Bhasha (Newari), an endangered language of the Kathmandu Valley, remains digitally marginalized due to the severe scarcity of annotated speech resources. In this work, we introduce Nw\=ach\=a Mun\=a, a newly curated 5.39-hour manually transcribed Devanagari speech corpus for Nepal Bhasha, and establish the first benchmark using script-preserving acoustic modeling. We investigate whether proximal cross-lingual transfer from a geographically and linguistically adjacent language (Nepali) can rival large-scale multilingual pretraining in an ultra-low-resource Automatic Speech Recognition (ASR) setting. Fine-tuning a Nepali Conformer model reduces the Character Error Rate (CER) from a 52.54% zero-shot baseline to 17.59% with data augmentation, effectively matching the performance of the multilingual Whisper-Small model despite utilizing significantly fewer parameters. Our findings demonstrate that proximal transfer from Nepali language serves as a computationally efficient alternative to massive multilingual models. We openly release the dataset and benchmarks to digitally enable the Newari community and foster further research in Nepal Bhasha.