README FILE FOR 2007 NIST Language Recognition Evaluation Supplemental Training Set
------------------------------------------

LDC Catalog-ID:  LDC2009S05


This DVD contains audio data and documentation that have been prepared
specifically for the 2007 NIST Language Recognition Evaluation
(LRE07).

The audio files contained here (under the "audio" directory) are
single-channel, 8kHz, ulaw, SPHERE-formatted.  Each file represents
one side of a 2-channel telephone conversation.  The files have been
subdivided according to the language used in each conversation.

Whereas the LRE07 Evaluation Plan involves a large number of distinct
target languages and dialects, this training release only provides
audio data in those languages for which no training data are available
in existing LRE-related corpora.  The languages provided here are:

  ABBREV. Full Name
  -------------------
    ARB   	  Arabic
    BEN	  Bengali
    CFR	  Min Nan Chinese
    RUS	  Russian
    THA	  Thai
    URD	  Urdu
    WUU	  Wu Chinese
    YUH	  Cantonese Chinese

Directory and file names are all in lower-case; each language
abbreviation is used as a directory name within the "audio" directory,
and individual data file names follow the pattern:

   lre07_tr_{lng}_{nnn}_{c}.sph

where:

   {lng} is the 3-letter language abbreviation
   {nnn} is a 3-digit sequential call-ID number
   {c} is the single-letter channel-ID ("a" or "b")

Note that if two files have the same call-ID number, they represent
the two sides of a single conversation.  In all (across all eight
languages), 212 distinct conversations are represented, 108 of which
have both sides presented in this release (104 conversations have only
one side present).  In all cases, the each audio file contains a
distinct, unique speaker.

The "docs" directory contains a table (call_side_info.csv) that
provides basic demographic data about each call side.  The
comma-delimited fields in that table are:

  LNG -- Language spoken (also the audio subdirectory name)
  CID -- 3-digit numeric call-ID (part of the file name)
  CHN -- Channel ("a" or "b", also part of the file name)
  SEX -- Speaker gender ("M" or "F")
  AGE -- Speaker age in years
  PHN -- Type of phone used ("L" = land-line, "C" = cell)

There are 40 call sides for each of the eight languages, so there are
320 rows of data in the table (plus an initial line providing the
column headings).  Age information is unavailable for some subjects.