Bemba Speech Translation: Exploring a Low-Resource African Language

TL;DR

This paper presents a speech translation system for Bemba, a low-resource African language, utilizing cascaded models, data augmentation, and synthetic data to improve translation quality.

Contribution

It introduces a novel approach combining Whisper and NLLB-200 models with data augmentation techniques for Bemba speech translation.

Findings

Synthetic data improves translation accuracy

Data augmentation enhances low-resource language translation

Cascaded systems outperform baseline models

Abstract

This paper describes our system submission to the International Conference on Spoken Language Translation (IWSLT 2025), low-resource languages track, namely for Bemba-to-English speech translation. We built cascaded speech translation systems based on Whisper and NLLB-200, and employed data augmentation techniques, such as back-translation. We investigate the effect of using synthetic data and discuss our experimental setup.