notesum.ai

Published at November 27

Aligning Pre-trained Models for Spoken Language Translation

cs.CL

cs.AI

cs.LG

Released Date: November 27, 2024

Authors: Šimon Sedláček, Santosh Kesiraju¹, Alexander Polok¹, Jan Černocký¹

Aff.: ¹Brno University of Technology

Arxiv: http://arxiv.org/abs/2411.18294v1

Refer to caption

Arch.	C	L	#P	How2 BLEU		How2 chrF2
Arch.	C	L	#P	val	test	val	test
ECD	Q	6	9.6M	44.0	43.9	65.0	64.8
	STE	6	10.7M	45.0	44.8	66.0	65.8
	STE	4	7.9M	44.1	44.4	65.3	65.3
ECED	Q	6	9.6M	44.1	44.2	65.3	65.2
	STE	6	10.7M	44.7	44.8	65.8	65.9
E-Branch. E2E			38.5M	45.6	45.2	66.6	66.1
Cascade			-	40.9	40.4	64.4	64.1