SPREDS-D1(SPeech Recognition Evaluation Data Set - Discourse type 1) ver1.2

What's new

• 2022/12/2 ver1.0 (Japanese) has been released.
• 2023/1/31 ver1.1 (Japanese) has been released. An unnecessary segment has been deleted.
• 2023/8/18 ver1.2 has been released. English (en) has been added.

Overview

SPREDS-D1 is a set of evaluation data for speech recognition released by NICT under the Creative Commons Attribution 4.0 International License (CC BY 4.0), consisting of long-duration speech of lectures, conferences, etc. by multiple people. The target languages are Japanese and English. The target domain is business and the data set contains recorded audio data of imaginary business meetings including presentations held by 1-3 persons, and their transcriptions. Tags used at NICT are included in the transcription data. For further details, please refer to '00README.txt' in each directory.

Download

ver1.2(Japanese)

 ver1.2(English) 

Extracted directory

The files have been compressed in 'xz' format. The extracted directory should look like the following. Please refer to '00README.txt' for further details about the files in 'LABEL' and 'WAV'.

-------------------------------------------------------------------------------------------
$ver =[version number]
$lang={ja,en}

doc/
$lang/
  00README.txt
  individual/
    unsegmented/
       LABEL/
       WAVE/
    segmented/
       LABEL/
       WAVE/
    mixed/
       LABEL/
       WAVE/
-------------------------------------------------------------------------------------------