en

Interspeech2020口音英语竞赛-数据堂

Organizers

  • Interspeech2020口音英语竞赛-数据堂-联合主办单位

    CCF Task Force on

    Speech Dialogue

    and Auditory Processing

  • Interspeech2020口音英语竞赛-数据堂-联合主办单位

    SHAANXI PROVINCIAL KEY LABORATORY

    OF SPEECH & LMAGE

    INFORMATION PROCESSING

  • Interspeech2020口音英语竞赛-数据堂-联合主办单位

    xi'an software park

  • Interspeech2020口音英语竞赛-数据堂-联合主办单位

    Shanxi Kunpeng 

    Ecological

    Innovation Center

  • Interspeech2020口音英语竞赛-数据堂-联合主办单位

    Datatang (Beijing) Techn

    ology Co., Ltd.

  • COMPETITION BACKGROUND

    INTERSPEECH 2020

    Interspeech, organized by ISCA is the world's largest and most comprehensive conference on the spoken language science. The global speech researchers, artificial intelligence enterprises in the field will conduct in-depth discussions here.

    As the flagship technical activity, the Accented English Automatic Speech Recognition Workshop will be held On October 25, 2020 in Shanghai. Our award ceremony will be held during the workshop.

  • INTRODUCTION OF COMPETITION

    About the Competition

    English is the most influential language in the world, and the English speech recognition system has achieved good effect at present. However, there are still some difficulties in recognizing people in heavy accents. In fact, the difficulties mainly stem from the inconsistent accent, the speech speed, etc. In addition, the shortage of accented English speech data has alsolimited the research.

    Thus the competition set up the following two subtasks:Track1 Accent Recognition, determining which country the speaker is from; Track2 English Speech Recognition, evaluating the speech recognition accuracy.

    Resources provided by Huawei Cloud, DATATANG.

TRACK SETTING

Track1

Accent Recognition

determining which country

the speaker is from

Track2

English Speech Recognition

evaluating the speech

recognition accuracy

Specified data

200 Hours-10 Different Countries Speakers English Speech Data

Collected by Mobile Telephone

525 speakers from ten countries participated in the recording of English speaking voice through Apple or Android phones. We collected 20 hours audio data from per country and the ratio of men and women in each country is 1:1. The recording environment is conducted in a relatively quiet room without echo. The recorded text covers many categories such as home furnishing, car-carrying, human computer interaction, etc. So the speech data has a rich content and wide fields, and the voice phonemes are also quite balanced.

Format

Speech data:16kHz,16bit,wav,mono

Annotation:txt

Data label:metadata

Recording

Environment

The speech data was recorded in relatively quiet room without echo

Recording

content

450 sentences per person

The actual recording and annotation process will discard some unqualified sentences, thus the final data

product may be less than 450 sentences per person

Corpus Types:

General language data: Sentences with unlimited fields and wide sources, including daily spoken language,

news and other content

Interactive speech data: involves different categories such as music, weather, travel, life, etc

Home commands data: involves control commands for smart home devices

Car commands data: related to the control commands of the equipment in car

Numbers: text data related to numbers, such as date, currency, time, etc

The average repetition of the recorded sentences is less than 3 times

Speaker

525 speakers from ten countries participated in the recording of English speaking voice; 20 hours of speech data per country; 50% women and 50% men

Countries: Russia, South Korea, Canada, United States, Portugal, Japan, Spain, India, United Kingdom, China

The voice data recorded by non-native English speakers includes the accent data and the non-accent data

Devices

Recording by Apple or Android phones

Mobile phones: Android phones, Apple phones, covering mainstream models of common brands on the market, such as Samsung, Huawei, Xiaomi, etc

Language

All the speakers use English for the recording

Annotate

content

The translated text was based on actual pronunciation of the audio

Accuracy

The sentence error rate (SER) is less than 5%

Competition Schedule

Awards

Note:All the prize amounts include the tax.

Participants

Open to the whole society such as colleges, scientific research institutes, Internet companies and other personnel can register for the competition.

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

Registration for the contest

  • Send the application form to [email protected]
  • with the subject [Accent English Contest - Team Name].
Download

Anti-cheating Statement

  • Participants are forbidden to submit multiple applications, and the results will be cancelled.

  • Participants are prohibited from using any other ways that outside of the designated assessment such as loopholes in the rules or technical loopholes, additional data or other undesirable ways to improve the ranking of results. Once found, the results will be cancelled.

All rights reserved by Data Palace (Beijing) Technology Co.

SOLUTIONS

Please fill in your name

Mobile phone format error

Please enter the phone number

Please fill in the full name of the company

Please fill in your e-mail

Requirement description cannot be empty!

Successful submission! Thank you for your support.

Format error, Please fill in again

Confirm

Minimum 5 characters required!

No data available

Terms Privacy Datatang. All Rights Reserved. Legal statement and privacy policy

*Name:

*Phone:

*Company:

*E-mail:

*Requirement:

数据堂_datatang