ACCENTED ENGLISH AUTOMATIC SPEECH RECOGNITION WORKSHOP & CHALLENGE

口音英语语音识别挑战赛

ACCENTED ENGLISH AUTOMATIC SPEECH RECOGNITION CHALLENGE 2020

Organizers

CCF Task Force on

Speech Dialogue

and Auditory Processing
Audio, Speech and Language

Processing Group, Northwestern

Polytechnical University
xi'an software park
SHAANXI KUNPENG

Ecological

Innovation Center
Speech Lab, Shanghai Jiao

Tong University, China
School of Computer Science and

Engineering, Nanyang Technological

University, Singapore
Center for Language and Speech

Processing, John Hopkins

University, United States
Datatang (Beijing)

Technology Co., Ltd.

CHALLENGE BACKGROUND

INTERSPEECH2020 Accented English Speech Recognition Workshop

INTERSPEECH has grown into the world's largest technical conference focused on speech processing and application. The conferences emphasize interdisciplinary approaches addressing all aspects of speech science and technology, ranging from basic theories to advanced applications. As a flagship sattelite workshop, the Accented English Speech Recognition Challenge (AESRC) will provide a common testbed for researchers in speech recogntion, especially recognition of speech with accents, and the challenge workshop will take place virtually.

CHALLENGE INTRODUCTION

Accented English Speech Recognition Chanllenge

English is the most influential universal language in the world. English speech recognition is also one of the most concerned areas in both academia and industry. At present, advanced ASR systems have achieved good effect and meet most requirements for standard English. In accent English field, however, recognizing English speech with accents still remains a challenging task. The difficulties in building an accent English ASR system mainly arise from the diversity of pronunciation accuracy, intonation speed and pronunciation of some syllables. On the other hand, the shortage of accent speech data limits the relevant research.

The Interspeech 2020 Accented English Speech Recognition Challenge (AESRC) will open 8 sets of accented English data from different countries to the participants, covering various pronunciation characteristics and accents, aiming to promote the discussion and exchange on English language research and accent speech recognition. It is expected that all researchers from academia and industry can learn from each other and truly gain by participating our challenge & workshop.

Computing resources will be provided by Huawei

TRACK SETTING

Track1

Accent Identification

Use permitted data only to train an accent recognition model. Submit the result of language identification on the test set.

Note：No limit for models and training technics. Evaluation considers the identification accuracy of the test set only.

Track2

Accented English Speech Recognition

Use permitted data only to train an ASR model for recognizing all kinds of accented English. Submit the result text of recognition.

Note: Test sets will include accents beyond training data in order to evaluate the generalization performance of the model. All kinds of system combination methods including ROVER are strictly prohibited. Language model training should only use the transcripts of permitted speech training data. Data augmentation should only be applied on the permitted speech data only.

Specified data

160 hours of labelled accented speech collected in Russia, Korea, US, Portugal, Japan, India, UK and China (20 hours/country) will be released to the registered teams.

Duration	20 hours ×8
Language & Accent	Accented English from Russia, Korea, US, Portugal, Japan, India, UK, China
Speaker	40 – 110 speakers per accent
Audio Format	16kHz, 16bit, single channel wav
Recording environment	Indoor, mobile phone
Speech content	Daily communication, interaction with smart devices, etc

Datasets will be released with metadata files organized in the following format

FIELD	DESCRIPTION
SEX	Speaker gender
AGE	Speaker age
ACT	Accent type
MIT	Recording device
SCC	Recording environment
LBR	Utterance duration
ORS	Raw text

Librispeech data is also permitted to use in both tracks. （http://www.openslr.org/12/）

Challenge Schedule

1
Registration

deadline
Aug 31, 2020
Aug 31, 2020
Release training

dataset
2
3
Release test sets

Sep 22, 2020
Sep 24, 2020
Submit results for

both tracks
4
5
Announce challenge

results
Sep 30, 2020
Oct 09, 2020
Submit system

description
6
7
Technical Seminar

& Award Ceremony(Online)
Dec 05,2020

Awards

Accent Recognition

1 First Prize - ¥ 10,000

2 Second Prizes - ¥ 5,000 each

3 Third Prizes - ¥ 2,000 each
English Speech Recognition

1 First Prize - ¥ 10,000

2 Second Prizes - ¥ 5,000 each

3 Third Prizes - ¥ 2,000 each

Note:All the prize amounts include the tax.

International Scientific Committee

(Names listed in no particular order)

Lei Xie	Northwestern Polytechnical University, China
Yanmin Qian	Shanghai Jiao Tong University, China
Shinji Watanabe	John Hopkins University, United States
Chng Eng Siong	Nanyang Technological University, Singapore
Qiangze Feng	Datatang（Beijing）Technology Co.,Ltd, China

Participants

Challenge is open to university, scientific research institutes, and internet enterprises.

Note: The challenge organizers and technical support units such as the employees who have the access to the business, products and data about the challenge will automatically withdraw from the challenge and give up the qualifications.

Registration

If you are interested in the challenge, please contact us by email to [email protected]
Download the registration form (either English or Chinese version), fill in the information, and send it to the email address above. The registration deadline is Aug 31 2020.
The organizing committee will review and verify the qualifications of the participating teams within 5 working days. The teams that have passed the review will sign the challenge data usage agreement, and qualified to join the challenge.
The training data will be announced on Aug 31 2020, and the data downloading method will be provided to the participants who have passed the review and signed the agreement.

Download

Evaluation and ranking

Note：

Team B, I, U2, K2, and M2 only submitted Track2 results. Team D2 and O3 only submitted Track1 results. Team Q3's submission was invalid.

2020AESRC Test Sets

Download test sets

Anti-cheating Statement

Participants are prohibited to register more than one time.
Participants are supposed to obey the data using rules of each track strictly. Teams that break the rules will be disqualified to use the data and the results will be invalid.

Q&A

Download FAQ for 2020AESRC