What question did this study set out to answer?

This study aims to evaluate speech-to-text translation for the endangered Dayak Ma’anyan language using advanced technology.

Mixed approach speech-to-text translation for endangered language

View Full Paper

Ask AI

Bookmark

Authors

Randomized trial evaluates speech-to-text translation for Dayak Ma’anyan, suggesting effective multilingual solutions.

This study aims to evaluate speech-to-text translation for the endangered Dayak Ma’anyan language using advanced technology.
Employs a cascaded speech-to-text translation framework using Whisper Large-v3 and SeamlessM4T v2.
Combines multilingual ASR models with LoRA-adapted translation models NLLB-200 3.3B and distilled 600M.
Tests conducted in a low-resource setting with limited parallel speech and text data.
Whisper achieved a lower word error rate (0.464) compared to SeamlessM4T (0.812), indicating superior ASR performance.
LoRA-adapted NLLB-200 3.3B achieved BLEU 31.00 and chrF 58.91, demonstrating the best machine translation performance.
ASR quality significantly influenced overall translation performance, with larger MT models showing better robustness against noisy outputs.

Sinaga et al. (2026) studied this question.

View Full Paper

Ask AI

Bookmark