Recent Activities for Speech IO Assessment and Speech Database in Korea YongJu Lee Director, Prof. S

1
Recent Activities for Speech I/O Assessment and
Speech Database in Korea Yong-Ju Lee
(Director, Prof./ SiTEC, Wonkwang University,
Korea)
2

• Creation Distribution of Language
  Resources for Common Use
• SiTEC (Speech Information Technology Industry
  Promotion Center)
• ETRI (Electronics Telecommunications Research
  Institute)

3
SiTEC(1)

• Dialogue corpus
• large speech corpus based on a large dialogue
  scenario which it generated in cooperation with
  ETRI for TV guide and telematics domain.
• 597 dialogue scenarios were made up taking the
  dialogue complexity into consideration for each
  of TV guide, weather, schedule, and telematics
  domains.
• And the dialogue corpus also contains alternative
  natural dialogues with alternative expressions
  having meanings equivalent to the expressions in
  the corresponding dialogues.
• Pieces of information on pragmatics, concept,
  domain, sentence regularization, discourse, and
  frame are tagged to 500 dialogues among 597
  scenarios.
• A dialogue speech corpus has been created of
  32,000 utterances made by 100 speakers according
  to these scenarios.

4
SiTEC(2)

• LILA project
• Creation of a number of spoken databases for
  training Automatic Speech Recognition Systems in
  the Asian Pacific area.
• Speech Data are collected through the mobile
  telephone network.
• The LILA consortium is composed of a large number
  of industrial companies. Each company is in
  charge of the production of a database. The
  consortium shares the databases produced in the
  project.
• ELRA is the key organization of this project.
• Together with HMT (a company in Korea) affiliated
  with SITEC, SITEC is working on creating a Korean
  speech corpus for the project of the LILA
  consortium assisted by ELRA.
• For this project, each of 1,000 speakers utters
  60 words and sentences through digital telephone
  lines. This project will be completed at the end
  of this year.

5

• Speech corpus for robot application
• For the implementation of  speech recognition in
  the robot, a corpus of remote speech by distance
  and direction has been created by using
  multi-channel microphone array and HAT( Head and
  Torso simulator)
• 400 speakers were recorded.  

6
ETRI

• ETRI has been constantly working on speech and
  text corpora as part of work on information
  resources management.
• This year ETRI has worked on creating speech
  corpora of foreign languages, such as English,
  Japanese, and Chinese, acquired in the cars and
  multi-channel environments, each of which
  contains 50,000 utterances.

7
Speech I/O assessment

• Researchers at ETRI and SITEC have just started
  to study on the procedures of objective
  assessment for performance of
• speech recognition in the robot
• and dialogue style TTS
• Hope to report on the details at the next
  meeting!!