#       FILE: release_v4.2.txt
#
#  DESCRIBES: Field information for the 
#             1997 Speaker Recognition Keys
#
#	      MALE ----  mal97_release_v4.2
#             FEMALE --  fem97_release_v4.2
#

Each of these keys contain 23 columns (fields).

Column | Description
----------------------
   1   |  Test segment name
-------|--------------
   2   |  Nominal test duration
-------|--------------
   3   |  Actual duration of the test-segment
-------|--------------
   4   |  na - will contain the conversation name as it will
       |       appear in the swbd-ii corpus.  This name will
       |       match the name in the test segment's header.
-------|--------------
   5   |  Speaker ID of the speaker in the test segment
-------|--------------
   6   |  "-" symbol not used currently, included to allow
       |  old scripts to parse the key properly
-------|--------------
   7   |  Channel of the conversation, from which the
       |  test segment was taken.
-------|--------------
   8   |  Gender "f" --> Female
       |         "m" --> Male
-------|--------------
   9   |  Indicates whether or not the test segment came from 
       |  a target or a non-target speaker.
       |    tar --> Target Speaker  (Training data provided)
       |    non --> Non-Target Speaker (No training data provided)
       |    XXX --> We are using as non-targets.  This is a case
       |            where we originally had the segment tagged as
       |            a target speaker, but then later found it to 
       |            be a different speaker, but we don't know who.
-------|--------------
  10   |  Indicates the relationship between the LINE used in
       |  the test segment in regards to the LINE(S) used for the 
       |  various training conditions.
       |      "-" -----> Non-Target Speaker
       |      "1h:" ---> This test segment came from a conversation
       |                 using the SAME LINE as was used in both
       |                 1-Handset & 1-Session training.
       |      "2h:" ---> This test segment came from a conversation
       |                 using the SAME LINE as was used in the 
       |                 2-Handset Training.
       |      "OUT" ---> This test segment came from a conversation by
       |                 a target speaker, but from a LINE that was
       |                 NOT used in any training conditions.
-------|--------------
  11   |  Indicates "noisiness" of the test-segment.
       |      "noise" ---> Some type of noise exist in the segment.
       |      "-----" ---> Basically clean
       |    
       |      NOTE: A noisy segment was determined by listening
       |      to the 10 second segment, but was included for the 
       |      corresponding 30 and 3 second segments.
       |
       |      ALSO NOTE: The noise is a binary decision which covered
       |      line noise (hum, static, speaker-phones...) as well as
       |      background noise (people talking, babies crying, radio/tv)
-------|--------------
  12   |  The amount of speech the speech detector found on the channel
       |  the tests segment was take from, for the period the test 
       |  segment was composed from.  
       |
       |  When used with field 13, you may determine one side of the
       |  conversation was being dominated by a certain channel, and
       |  it may give insight to the "type" of speech being offered
       |  by the target.
-------|--------------
  13   |  Same as 12 but the speech is for the other channel, who
       |  the target is speaking to.
-------|--------------
  14   |  Age of speaker
-------|--------------
  15   |  Dialect of target speaker.
       |          Western        WS
       |          Foreign        FR
       |          Northern       NO
       |          North Midland  NM
       |          South Midland  SM
       |          Southern       SO
       |          New England    NE    
       |          New York City  NY  
       |          Mixed          MX
       |          Unknown        --
-------|--------------
  16   |  Average pitch of the "hs1_s1a" training segment segment (Hz)
-------|--------------
  17   |  Average pitch of the "hs1_s1b" training segment segment (Hz)
-------|--------------
  18   |  Average pitch of the "hs1_s2" training segment segment (Hz)
-------|--------------
  19   |  Average pitch of the "hs2" training segment segment (Hz)
-------|--------------
  20   |  Average pitch of the 30 second test segment segment (Hz)
       |
       |  NOTE this value was not transferred to the 
       |  10 and 3 second segments.
-------|--------------
  21   |  Number of calls the speaker participated in, for the 
       |  SWITCHBOARD-II phase 1 corpus.
-------|--------------
  22   |  Handset type of the Test segment
-------|--------------
  23   |  Handset type(s) of the 1-Session Training Data
       |     CARB -> If both were found to be carbon
       |     ELEC -> If both were found to be electret
       |     MIX --> If one training minute was found to be
       |             carbon, and the other minute electret.
----------------------
