2016年6月2日 (四) 07:21的版本

Oriental Multilingual Speech Processing and Oriental Language Recognition (OLR) Challenge

• Organizers: Dong Wang (Tsinghua Univ.), Qing Chen (Speechocean)
• Email: wangdong99@mails.tsinghua.edu.cn; chenqing@speechocean.com

Introduction

Oriental languages involve some common properties as well as sufficient diversity in acoustic, phonetic, lexical and linguistic aspects, leading to interesting questions such as which languages share more and which languages are more discriminant, and how the commonality and diversity can be used to assist speech processing research. This special session focuses on multilingual speech processing for oriental languages, including but not limited to phonetic/phonological comparative study, multilingual speech recognition, multilingual speech synthesis. To support the research, this special session will release an oriental multilingual speech database THT-OL10 (provided by Speechocean) which involves 10 oriental languages (including Australia English). Based on this database (free for session participants), we call for an oriental language recognition (OLR) challenge.

Target track

Speech and language processing

Scope

This special session is expected to attract papers on recent research progress in the area of multilingual speech processing. The targeted research topics are, but not limited to, the following:

•  Multilingual phonetic and phonological analysis
•  Multilingual speech recognition
•  Multilingual speech synthesis
•  Multilingual speaker recognition
•  Multilingual language understanding

OLR Challenge

The THT-OL10 database involves 10 oriental languages, each consisting of 20 hours of speech signals. This special session calls for a OLR challenge which discriminates the 10 languages of THT-OL10. The data will be free to institutes who (1) participate the OLR challenge; (2) do not participate OLR but participate this special session, and require data to evaluate their research.