notesum.ai

Published at November 27

Continual Learning in Machine Speech Chain Using Gradient Episodic Memory

cs.CL

cs.AI

eess.AS

Released Date: November 27, 2024

Authors: Geoffrey Tyndall¹, Kurniawati Azizah¹, Dipta Tanaya¹, Ayu Purwarianti², Dessi Puji Lestari², Sakriani Sakti³

Aff.: ¹University of Indonesia; ²Bandung Institute of Technology; ³Nara Institute of Science and Technology

Arxiv: http://arxiv.org/abs/2411.18320v1

Refer to caption

Model	CER (%) LJ Original	CER (%) LJ Noisy
ASR ${}_{\text{Lower}}$
Pre-trained	9.2	82.6
Fine-tuning	19.0	31.3
GEM	8.5	15.8
Multitask	74.8	76.7
ASR ${}_{\text{SpeechChain}}$
Pre-trained	6.4	95.7
Fine-tuning	12.7	33.1
GEM	11.1	15.5
ASR ${}_{\text{Upper}}$
Pre-trained	1.9	108.4
Fine-tuning	6.7	15.6
GEM	5.2	8.4
Multitask	3.8	10.9