homink/speech.ko

Korean read speech corpus (about 120 hours, 17GB) from National Institute of Korean Language

speech.ko

Korean read speech corpus (about 120 hours, 17GB) from National Institute of Korean Language (NIKL)

This repository cleans up NIKL corpus such as voiding unnecessary wav files, matching sampling rate, trimming silence, etc. Details are below.

Location

http://www.korean.go.kr/front/board/boardStandardView.do?board_id=4&mn_id=17&b_seq=464

https://ithub.korean.go.kr/user/corpus/referenceManager.do

Once you download all corpus files from above, you will find the following zip files.

ls *.zip
3-1 #1 (20대 남성) 4-1.zip  3-2 #1 (30대 남성) 5-1.zip  3-2 #2 (40대 여성) 5-5.zip
3-1 #1 (20대 남성) 4-2.zip  3-2 #1 (30대 남성) 5-2.zip  3-3 #3 (50대 이상 남성여성) 6-1.zip
3-1 #1 (20대 남성) 4-3.zip  3-2 #1 (30대 남성) 5-3.zip  3-3 #3 (50대 이상 남성여성) 6-2.zip
3-1 #1 (20대 남성) 4-4.zip  3-2 #1 (30대 남성) 5-4.zip  3-3 #3 (50대 이상 남성여성) 6-3.zip
3-1 #2 (20대 여성) 5-1.zip  3-2 #1 (30대 남성) 5-5.zip  3-3 #3 (50대 이상 남성여성) 6-4.zip
3-1 #2 (20대 여성) 5-2.zip  3-2 #2 (40대 여성) 5-1.zip  3-3 #3 (50대 이상 남성여성) 6-5.zip
3-1 #2 (20대 여성) 5-3.zip  3-2 #2 (40대 여성) 5-2.zip  3-3 #3 (50대 이상 남성여성) 6-6.zip
3-1 #2 (20대 여성) 5-4.zip  3-2 #2 (40대 여성) 5-3.zip
3-1 #2 (20대 여성) 5-5.zip  3-2 #2 (40대 여성) 5-4.zip

Transcription file contains 19 topics with the following number of sentences.

Topic	# of sentence
1	51
2	87
3	69
4	62
5	47
6	54
7	62
8	94
9	60
10	73
11	42
12	28
13	39
14	27
15	17
16	35
17	19
18	27
19	40
Total	930

When unzipped, you will find speaker ids as follows.

Gender	Speaker ID	Age
Female	fv01 to fv20	20s
Female	fx01 to fx20	40s
Female	fy01 to fy18	Older than 50s
Female	fz05 to fz06	Older than 50s
Male	mv01 to mv20	20s
Male	mw01 to mw20	30s
Male	my01 to my11	Older than 50s
Male	mz01 to mz09	Older than 50s

Dependencies

sox

sox is used to check wav file info such as sampling rate and correct format, and used to convert sampling rate if necessary.

auditok

auditok is used to trim unnecessary silence in the beginning and in the end.

Command

git clone https://github.com/homink/speech.ko.git
cd speech.ko
./run.sh --corpus ${corpus_location}

Cleaned and trimmed wav files will be found in ${corpus_location}/trimmed_data.

speech.ko's Issues

Could you please share the transcript files

Thank for your sharing.

There are problems with your respository, so I cannot run the run.sh file.

Anyway, I can download the file from home page (the links you have shared)
But I could not find the transcript files. Could you please share them to me? Or give me download links?

I appreciate your helps. Thank you so much..

There may be some changes in database.

Thank you for making this tools.

There is a question. When I downloaded the database from NIKL, I found that there were 87453 files but, your script check that there are 87407.

Q1. Is there as version difference in database?
Q2. If the database changed, how can I fix it?

Cannot download files

I see the following error when I click on the first link to download the files:

 서비스 이용에 불편을 드려 죄송합니다.

서버 프로그램에서 예상치 못한 오류가 발생하여 해당 페이지를 열람할 수 없습니다.
자세한 사항은 홈페이지 관리자에게 문의 바랍니다.