TITLE:    Levantine Arabic QT Training Data Set 5 (Speech + Transcripts)
		     [LDC2006T07] 
	  

Authors: Mohamed Maamouri (Project head), Tim Buckwalter, Dave Graff, Hubert Jin

This release contains 1660 calls and the total speech is approximately 250
hours of telephone conversation in Levantine Arabic. 

In this directory (docs), we included the following documents:

   1) filelist

        A list of conversation IDs with prefix of 'fsa_'.

   2) wordlist.utf8.txt

        Wordlist mapping table

   3) speaker_info.txt

	Speaker information on origin, gender, age (group) etc, judged
        by the annotators who transcribed the conversations.

   4) fla_pindata.tbl

	Speaker PINs and phone numbers for each call in the corpus. This
        information is from registration of the speakers. Although it
        provides important information to group speech from the same 
        speaker, it is not always the case that the person who talked 
        in the call was the one who registered.

This corpus is the combination of 4 former training data sets: LDC2004E22,
LDC2004E66, LDC2005T03 and LDC2005S14(Speech and Transcripts). More than half of the speakers are
Lebanese.  The break down of the dialects is as follows:

 559 JOR
1853 LEB
 355 PAL
  67 SYR

 484 Lev (general category of Levantine Arabic)


Directory structure

    annotation - 1660 transcription files in the UTF-8 format.
    audio      - 1660 audio files in the sphere format.
    docs       - documentation.

    Note: sph2pipe can be used to convert the sphere files to wave files.
          For more information, please search refer to http://www.ldc.upenn.edu/Using.
    
For more information, please contact

    Mohamed Maamouri     maamouri@ldc.upenn.edu
    Timbuck Water        timbuck2@ldc.upenn.edu
    Dave Graff	         graff@ldc.upenn.edu
    Hubert  Jin	         hubertj@ldc.upenn.edu