Tied multitask learning for neural speech translation

A Anastasopoulos, D Chiang - ar**s from 218 allophones to
phonemes for 14 languages. Phonemes are contrastive phonological units, and allophones …

[BOG][B] Computational tools for endangered language documentation

A Anastasopoulos - 2019 - search.proquest.com
COMPUTATIONAL TOOLS FOR ENDANGERED LANGUAGE DOCUMENTATION A
Dissertation Submitted to the Graduate School of the University of N Page 1 …

AlloST: low-resource speech translation without source transcription

YF Cheng, HS Lee, HM Wang - arxiv preprint arxiv:2105.00171, 2021 - arxiv.org
The end-to-end architecture has made promising progress in speech translation (ST).
However, the ST task is still challenging under low-resource conditions. Most ST models …

A small Griko-Italian speech translation corpus

MZ Boito, A Anastasopoulos, M Lekakou… - arxiv preprint arxiv …, 2018 - arxiv.org
This paper presents an extension to a very low-resource parallel corpus collected in an
endangered language, Griko, making it useful for computational research. The corpus …