• The word time-stamps in the ASR transcripts for the 10-sec segments were not adjusted to make the segments start at time zero. Download the corrected CTM's (April 26, 2006 - v2).
  
  
• Test segment pdag.sph has a bad checksum in its header. Download the new pdag.sph with the corrected checksum in its header.
  
  
• Current version v4 - May 10, 2006: 8conv4w-1conv2w.ndx in v3 still has some trials with the channel designation of the test segments as 'B'. It is corrected in v4. Also, correcting this problem produces duplicate trials in all the 2w test conditions and these duplicate trials are removed in v4.
  
  
  • v3 - May 1, 2006: v2 contains three additional index files for the train-test conditions not listed in the evaluation plan. These extra index files (10sec4w-1conv2w, 10sec4w-1conv4w, 3conv2w-10sec4w) are removed in v3. Also in v2, some trials in the *-1conv2w test condition index files have the channel of the test segment as 'B' instead of 'A'. They are corrected in v3 to reflect that for the summed channel test, the channel of the test segment is always 'A'. 8conv4w-1conv2w.ndx in
    
    
  • v2 - April 27, 2006: Some index files in v1 have duplicate entries and trials referrring to training models of the incorrect type. We corrected these problems in v2.
    
    
  • v1 - April 24, 2006: Initial release.
    
    
• Reports of empty segments: NIST will verify these segments but sites should include such trials in their output.

[ SRE 2006 ]