notesum.ai

Published at November 29

Noro: A Noise-Robust One-shot Voice Conversion System with Hidden Speaker Representation Capabilities

cs.SD

cs.CL

eess.AS

Released Date: November 29, 2024

Authors: Haorui He¹, Yuchen Song¹, Yuancheng Wang¹, Haoyang Li², Xueyao Zhang¹, Li Wang¹, Gongping Huang³, Eng Siong Chng², Zhizheng Wu¹

Aff.: ¹School of Data Science, Chinese University of Hong Kong, Shenzhen 518172, China; ²School of Computer Science and Engineering, Nanyang Technological University, Singapore 639798, Singapore; ³School of Electronic Information, Wuhan University, Wuhan 430072, China

Arxiv: http://arxiv.org/pdf/2411.19770v1

Model	Clean Setting		Noisy Setting
	CER $\downarrow$	SECS $\uparrow$	CER $\downarrow$	SECS $\uparrow$
Ground Truth	1.37	90.4	1.37	90.4
DiffVC [1]	16.91	70.38	15.72	69.09
FreeVC [28]	3.31	80.07	3.49	73.83
FaCodec-VC [29]	3.66	82.33	3.56	74.78
Our Baseline	4.71	83.45	7.26	77.28
Noro	4.74	82.38	4.66	80.09