IFA Dialog Video corpus
R.J.J.H. van Son, Wesseling, Wieneke · arXiv (Cornell University) · 2008
IFA Dialog Video corpus All materials are licensed under the GNU general public license © 2007, Nederlandse Taalunie This corpus was made possible by grant 276-75-002 of the Netherlands Organization for Scientific Research Introduction The IFA Dialog Video corpus is a collection of annotated video recordings of friendly Face-to-Face dialogs. It is modelled on the Face-to-Face dialogs in the Spoken Dutch Corpus (CGN). The procedures and design of the corpus were adapted to make this corpus useful for other researchers of Dutch speech. For this corpus, 24 dialog conversations of 15 minutes were recorded, in total 6 hours of speech. To stay close to the very useful Face-to-Face dialogs in the CGN, pairs of well acquainted participants were selected, either good friends, relatives, or long-time colleagues. The participants were allowed to talk about any topic they wanted. In total, 20 of these conversations, 5 hours, were annotated to the same, or updated, standards as the original CGN. Only the initial orthographic transcription was done by hand. Other CGN-format annotations were only done automatically. Two other manual annotations were added, a functional annotation of dialog utterances and annotated gaze direction. Recordings are coded by camera (DVA or DVB), converstation number (1-24, recordings 5, 18, 21, and 23 were not annotated), and speaker ID (A-AP). For example, DVA16AA and DVB16AB are the two camera recordings made from conversation 16, with subject AA recorded on camera A and subject AB recorded on camera B. Some subjects participated in more than one conversation, always with different partners. See also the LREC paper The IFADV corpus: A free dialog video corpus(van Son, R., Wesseling, W., Sanders, E., and van den Heuvel, H. (2008). LREC'08, Marrakech) Other annotations can be found in El Haddad (2025). Recordings Recordings were made with two gen-locked JVC TK-C1480B analog color video cameras between April 25 and August 9, 2006. Specification: Image pickup: 1/2 type IT CCD 752 (H) x 582 (V) Synchronization: Internal Line Lock, Full Genlock Scanning frequency: (H) 15.625kHz x (V) 50Hz Resolution: 480 TV lines (H) A: Ernitec GA4V10NA-1/2 lens (4-10mm) B: Panasonic WV-LZ80/2 lens (6-12mm) Gen-lock ensures synchronization of all frames of the two cameras. Recordings were digitized using two Canopus ADVC110 digital video converters. Recordings were stored unprocessed on disk, ie, in DV format with 48 kHz 16 bit PCM sound. Each camera was positioned to the left of one speaker and focussed on the face of the other. Subjects wore a Samson QV head-set microphone. Subjects first spoke some scripted sentences. Then they were instructed to speak freely while preferably avoiding sensitive material or identifying people by name. All subject signed an informed consent and transfered all copyrights to the Dutch Language Union (Nederlandse Taalunie). Point your IMDI browser to the file 'Annotations/IMDI/IFADVcorpus.imdi' or: http://www.fon.hum.uva.nl/IFA-SpokenLanguageCorpora/IFADVcorpus/Annotations/IMDI/IFADVcorpus.imdi Materials Release note:The original recordings contained dropped frames which made the two recordings of each dialog become out-of-sync. This has been corrected by duplicating frames. This procedure is described in the SMILoverlay files. Only the corrected recordings are made available here. The original recordings, with the lacking frames, are available on request. Recordings are limited to 900 seconds (15 min) and corrected for dropped frames. That is, the video frames and audio files of both recordings are synchronized. Compressed video, with automatically normalized brightness and contrast levels AVI: MPEG-4 Part 2/DivX3, encoded video recordings (~287 MB). Also normalized for sound volume. mencoder -quiet -af volnorm=2:0.25 -vf pp=autolevels:fullyrange -of avi -ovc lavc -lavcopts vcodec=msmpeg4:vbitrate=2400000:vhq:keyint=50 -oac mp3lame -o outfile.avi infile.dv; OGV: Ogg Theora encoded video recordings (~200 MB) ffmpeg2theora --format dv --videoquality 4 --sharpness 1 --pp autolevels:fullyrange --license GPLv2 -o outfile.ogv infile.dv Speech files, extracted from the recordings.Extensions WAV: RIFF/WAV (uncompressed) FLAC: Free Lossless Audio Codec (compressed) Annotations of the stereo sound of the A recording (ie, subject A on the left channel B on the right channel)Labeling was done by Anita van Boxtel at SPEX under supervision of Eric Sanders and Henk van den Heuvel(see: van Son et al, 2008 for annotation labels) ort: Orthographic transcription (Praat TextGrid) awd: Automatic word and phoneme alignment of ort (Praat TextGrid) pos: Automatic Part-of-Speech labeling (CGN format) EAF: Elan annotations for gaze direction g: gazing at the partner, x: looking away, d: looking down, k: blinking dbl: Normalized combination of other annotations (Praat TextGrid) DVA*.ort.awd.dbl: Functional dialogue annotations DVA*.gaze.dbl: Gaze annotation DVA*.ainton.dbl: Automatically determined end intonation (1: low, 2: mid, 3: high) scripts: Maintenance scripts for annotations Transcripts: Readable dialog transcriptsCorresponding audio files can be found in the Speech files directory. Summaries of the dialogsCompiled by Stephanie Wagenaar IMDI files of the recordingsCompiled by Maaike van Naerssen (note that not all recordings are annotated) scriptsScripts used to record and process the dialogs SMILoverlaysSMIL xml files to correct the dropped frames in the recordings in DialogCorpus tablesMetadata on the recordings and the speakers as well as tab-separated-values database tables with a single record for every annotated item in the Annotations folder DocumentsForms and published papers. Annotation instructions (HTML)Annotation instructions (Praat Manual)Annotation instructions are in Dutch IFA Dialog Video Corpus.Copyright © 2007 Nederlandse Taal UnieThis corpus was made possible by grant 276-75-002 of the Netherlands Organization for Scientific Research'Integration of information in spoken communication'Created by R.J.J.H. van Son and Wieneke Wesseling of the ACLC. Annotations were performed by SPEX Please note that these materials are distributed under the the GNU Affero General Public License v3.0 or later license. This license only covers the Copyright protection of the corpus. These materials are shared under the expectation that they are used ethically and responsibly. Publishing or broadcasting of materials from this corpus might be covered by other laws, eg, laws protecting the privacy and "good name" of the subjects. This is especially relevant if the materials are used outside of an educational or R&D context. Please read the forms in the Documents directory for more information (in Dutch).