KIT's Low-resource Speech Translation Systems for IWSLT2025: System Enhancement with Synthetic Data and Model Regularization

TL;DR

This paper introduces system enhancements for low-resource speech translation using synthetic data, model regularization, and system combination techniques, leading to improved performance across multiple language pairs.

Contribution

We propose novel methods including synthetic data augmentation, intra-distillation, and system combination to improve low-resource speech translation systems.

Findings

Synthetic data improves translation quality for low-resource languages.

Intra-distillation enhances model performance across tasks.

Combining systems yields approximately 1.5 BLEU point improvement.

Abstract

This paper presents KIT's submissions to the IWSLT 2025 low-resource track. We develop both cascaded systems, consisting of Automatic Speech Recognition (ASR) and Machine Translation (MT) models, and end-to-end (E2E) Speech Translation (ST) systems for three language pairs: Bemba, North Levantine Arabic, and Tunisian Arabic into English. Building upon pre-trained models, we fine-tune our systems with different strategies to utilize resources efficiently. This study further explores system enhancement with synthetic data and model regularization. Specifically, we investigate MT-augmented ST by generating translations from ASR data using MT models. For North Levantine, which lacks parallel ST training data, a system trained solely on synthetic data slightly surpasses the cascaded system trained on real data. We also explore augmentation using text-to-speech models by generating synthetic speech from MT data, demonstrating the benefits of synthetic data in improving both ASR and ST performance for Bemba. Additionally, we apply intra-distillation to enhance model performance. Our experiments show that this approach consistently improves results across ASR, MT, and ST tasks, as well as across different pre-trained models. Finally, we apply Minimum Bayes Risk decoding to combine the cascaded and end-to-end systems, achieving an improvement of approximately 1.5 BLEU points.