Despite a narrowed performance gap with direct approaches, cascade solutions, involving automatic speech recognition (ASR) and machine translation (MT) are still largely employed in speech translation (ST). Direct approaches employing a single model to translate…