KIT's Multilingual Speech Translation System for IWSLT 2023
This addresses the problem of adapting speech translation systems to real-life conditions like accented speech and specialized content for multilingual applications, though it is incremental as it builds on existing methods like kNN-MT and adapters.
The paper tackled speech translation for accented and terminology-dense scientific conference talks into 10 languages, using a retrieval-based approach and adapters to adapt without target domain training data, achieving a +0.8 BLEU improvement and showing cascaded systems outperform end-to-end ones in this domain.
Many existing speech translation benchmarks focus on native-English speech in high-quality recording conditions, which often do not match the conditions in real-life use-cases. In this paper, we describe our speech translation system for the multilingual track of IWSLT 2023, which evaluates translation quality on scientific conference talks. The test condition features accented input speech and terminology-dense contents. The task requires translation into 10 languages of varying amounts of resources. In absence of training data from the target domain, we use a retrieval-based approach (kNN-MT) for effective adaptation (+0.8 BLEU for speech translation). We also use adapters to easily integrate incremental training data from data augmentation, and show that it matches the performance of re-training. We observe that cascaded systems are more easily adaptable towards specific target domains, due to their separate modules. Our cascaded speech system substantially outperforms its end-to-end counterpart on scientific talk translation, although their performance remains similar on TED talks.