View source

{{Dataset <!-- IMPORTANT: please complete the fields after the = mark; remove any unassigned parameters -->
| domain = Broadcast
| media = Audio & VIdeo
| size = 133 hours
| instances = 1
| format = mxf, mpeg, asf 
| created = 2011
| task = Speech recognition, shot segmentation, content summarization, copy detection
| copyright = EBU
| url = http://ebu-scaie.lab.vrt.be/mammie/
}}

== Description ==

The EBU-SCAIE content set is an audiovisual content set that has
been made available by the European Broadcasting Union (EBU). 
The collected content
in the set is consisting of broadcast media content collected
from different broadcasters around the world. This content set
is made available to the research community in order to evaluate
automatic information extraction tools on this broadcast media. 
The set also contains ground truth data for several automatic
information extraction tasks. 

The MIM-SCAIE content set contains broadcast media productions from 5 different broadcasters: Radiotelevisione Italiana (RAI, Italy), Vlaamse Radio- en Televisieomroeporganisatie (VRT, Belgium), Nippon Hoso Kyokai (NHK, Japan), Deutsche Welle (DW, Germany) provided by Institut für Rundfunktechnik (IRT, Germany) and Danish Radio (DR, Denmark). RAI, VRT, NHK and DW provided video files. The material from DR are high resolution audio files.

The set currently includes 274 items. Each item is available as a high resolution video file, together with high-res audio tracks and corresponding ground truth information (if available). In total, there is 133 hours of audiovisual media available in the content set.

The content set can be reached through its web plaform called mammie: http://ebu-scaie.lab.vrt.be/mammie/


== Quality ==

Video quality:
{| class="wikitable"
|-
! Broadcaster !! Wrapper !! codec !! resolution !! bitrate
|-
| IRT || mpeg || mpeg2   || 720x576 || 12,5Mbit
|-
| VRT || mxf  || dvvideo || 720x576 || 32Mbit
|-
| NHK || mpeg || mpeg1   || 352x240 || 1150kbit
|-
| DR  || n/a  || n/a     || n/a     || n/a
|-
| RAI || asf  || wmv2    || 384x288 || 338kbit
|-
|}

Audio quality:
{| class="wikitable"
|-
! Broadcaster !! codec !! samplerate !! bitrate
|-
| IRT || mp2   || 48000 Hz || 384 kb
|-
| VRT || pcm   || 48000 Hz || 768 kb
|-
| NHK || mp2   || 44100 Hz || 224 kb
|-
| DR  || mp2   || 48000 Hz || 384 kb
|-
| RAI || wmav2 || 44100 Hz|| 64 kb
|-
|}

== Source ==

The data is obtained from different broadcast organisations: IRT, VRT, NHK, DR and RAI. 



== Ground Truth Annotation ==

For a fraction of the content, there is also ground truth information available. All videos from NHK are associated 
with ground truth for speech recognition and item segmentation. Also the content of VRT has ground truth 
available for some videos: item segmentation (4 videos), face detection (2 videos) and speech recognition 
(1 video).

Available ground truth:
{| class="wikitable"
|-
! Task !! # videos
|-
| Item segmentation  || 48
|-
| Speech recognition || 45
|-
| Face detection     || 2
|-
|}


== Features ==

== Licensing / Copyright ==

[http://ebu-scaie.lab.vrt.be/mammie/p-scaie_019_agreement.pdf] License agreement