arXiv · 2611.13512
Nepali-English Code-Mixed Speech Recognition: Data Preparation and Adaptation of a Speech Foundation Model
Abstract
Nepali speakers regularly mix English words into conversation, but multilingual speech recognisers often struggle with language switches and the writing script used for English words. We study how to adapt SraVaani, a 449M-parameter hybrid TDT-CTC model, using a 172.8-hour corpus covering six Nepali speech domains. Its pretrained tokenizer does not support digits and several common symbols. Rather than replace it or discard these recordings, we generate Nepali and English readings for numbers and use zero-shot ASR output to select a plausible form. Full fine-tuning reduces test word error rate (WER) from 48.6% to 17.1%; a 12-epoch run reaches 16.3%, compared with 16.4% for fine-tuned Whisper large-v3-turbo. Freezing the encoder increases WER to 25.5%, while excluding numeric clips raises WER on numeric test speech from 14.3% to 21.3%. We also evaluate on FLEURS using both its original references and mixed-script alternatives. The comparison shows that transcription conventions explain part, but not all, of the out-of-domain performance loss.
Explore related subjects
Keep this discovery
Explore connections, maps & timelines
Jeevan Neupane, Janardan Bhetwal, Kshitiz Paudel, Anup Aryal, Bipin Bashyal. 2026-11-10. Nepali-English Code-Mixed Speech Recognition: Data Preparation and Adaptation of a Speech Foundation Model. https://searcharxiv.com/abs/2611.13512
Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite Searcharxiv by Indiciae: BibTeX · RIS.