This README file contains the instructions for using the "SSPNet Conflict Corpus".
If you use the data, please cite the following article:

S.Kim, M.Filippone, F.Valente and A.Vinciarelli, “Predicting the Conflict Level in
Television Political Debates: an Approach Based on Crowdsourcing, Nonverbal
Communication and Gaussian Processes“, Proceedings of ACM International
Conference on Multimedia, pp. 793-796, 2012.

For any questions, please contact Alessandro Vinciarelli (vincia@dcs.gla.ac.uk)

-----

The Corpus includes the following material:

- videodata.zip 
Content: Archive of the 1430 video clips of the Corpus
Format: zip archive including flv videos
Size: 2.67 GB 

- audiodata.zip (audio channel of the 1430 video clips of the Corpus)
Content: Audio channel of the 1430 Corpus clips
Format: zip archive including wav recordings
Size: 3.62 GB 

- participantslist.csv
Content: list of debate participants
Format: csv
Size: 13 kB 

- manualdiarization.zip
Content: Speaker segmentation of the Corpus clips (one file per clip)
Format: csv
Size: 372 kB 

- conflictlevel.csv
Content: conflict level of the Corpus clips (groundtruth)
Format: csv
Size: 32 kB 

-----

Naming conventions

The clips are named in the following way:

xx-yy-zz_start_end.aaa

where

xx 		year when the debate was televised (e.g., '06' corresponds to 2006)
yy 		month when the debate was televised (e.g., '04' corresponds to April)
zz 		day of the month when the debate was televised (e.g., '03' corresponds
		to the 3rd day of the month)
start		start time of the clip in the debate in seconds (e.g., 318 means that the
		clip starts at second 318 of the debate televised on day zz of mont yy in
		year zz).
end		end time of the clip in the debate in seconds (same convention as the
		start time)
aaa		extension ('wav' for audio and 'flv' for video)

All clips for which the prefix xx-yy-zz is the same belong to the same debate.
Clips for which the prefix xx-yy-zz is different belong to different debates.

-----

Format of "participantslist.csv'

Column 1:		speaker code in the format spk_nn (nn is an ID)
Column 2:		name of the speaker in the format "firstname lastname"
Column 3:		gender of the speaker (male or female)
Column 4:		type of speaker (participant or moderator)
Column 5:		debate where the participant appears in the format
				xx-yy-zz (see naming conventions above)

The columns after the fifth include further debates where the speaker
appears (if any).

-----

Format of manual diarization csv files

Diarization files are named using the same conventions as the clips (see above),
but have extension 'csv'

Column 1:		Start time of the turn
Column 2:		End time of the turn
Column 3:		speaker code (see format of participantslist.csv) 
Column 4:		speaker code (only in case of overlapping speech)

-----

Format of 'conflictlevel.csv'

Groundtruth associated to the clip:

Column 1:		Name of the clip in format xx-yy-zz_start_end.aaa
				(see naming conventions above)
Column 2:		Groundtruth conflict level (float number)