notesum.ai

Published at October 22

Enhancing Low-Resource ASR through Versatile TTS: Bridging the Data Gap

cs.NI

cs.AI

cs.CR

cs.LG

Released Date: October 22, 2024

Authors: Guanrou Yang¹, Fan Yu², Ziyang Ma¹, Zhihao Du², Zhifu Gao², Shiliang Zhang², Xie Chen¹

Aff.: ¹MoE Key Lab of Artificial Intelligence, AI Institute, X-LANCE Lab, Shanghai Jiao Tong University, China; ²Institute for Intelligent Computing, Alibaba Group, China

Arxiv: https://arxiv.org/abs/2410.16726v1

Data Type	Language	Real(h)	TTS(h)	WER /CER(%)	Rel Gain(%)
Accented English	English	300	0	12.61	15.23
Accented English	English	300	300	10.69	15.23
Minority Language	Korean	3700	0	8.48	25.59
Minority Language	Korean	3700	4000	6.31	25.59
Chinese Dialects	Min Nan	200	0	55.42	45.85
	Min Nan	200	200	30.00	45.85
	Wu	200	0	27.47	27.45
	Wu	200	200	19.93	27.45
Hotwords	Chinese	300	0	21.32	38.33
Hotwords	Chinese	300	300	13.15	38.33