MAGIS

Models for the Automatic Generation of Interlingual Subtitles

CINECA ISCRA-B PROJECT

Scalable computing: Scientific use cases that require highly parallel jobs on large scale HPC architectures

MAGIS aims to deploy speech-translation (ST) technology for automatic subtitling (AS). In particular, the focus is on 3 aspects:
a) enhancing direct ST architectures to perform the whole process (translation, segmentation, timestamps prediction) in an end-to-end fashion,
b) moving from controlled lab testing environments to realistic settings involving new and diverse types of data,
c) scaling to more languages to explore the potential of our methods also on less-resourced ones.