Blame view

egs/sre08/v1/local/det_curve_example.sh 1.83 KB
8dcb6dfcb   Yannick Estève   first commit
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
  #!/bin/bash
  
  # This shell-script is just to show how you would plot det curves
  # using DETWare.
  
  # This assumes you have a file called "foo" with the scores in.
  
  # Note: there are some comments at the bottom showing what you
  # have to run in matlab.  It won't get run if you just run the script,
  # you have to do it manually.
  
  trials=data/sre08_trials/short2-short3-female.trials
  
  for condition in $(seq 8); do
  (
    # see http://www.itl.nist.gov/iad/mig/tests/sre/2008/official_results/
    # for interpretation of condition here,  e.g. 1 = "Interview train and test".
    # Condition 6 is "telephone train and test".
    # The EER is 
    awk '{print $3}' foo | paste - $trials | awk -v c=$condition '{n=4+c; if ($n == "Y") print $1, $4}' | grep -w target | \
      awk 'BEGIN {printf( "target = [ " );} {print $1} END{printf("];
  ");}'
    awk '{print $3}' foo | paste - $trials | awk -v c=$condition '{n=4+c; if ($n == "Y") print $1, $4}' | grep -w nontarget | \
      awk 'BEGIN {printf( "nontarget = [ " );} {print $1} END{printf("];
  ");}'
   ) > ~/DETware_v2.1/scores${condition}.m
  done
  
  # Note: the DETware_v2.1 directory is as extracted from the DETware package,
  # which I got at.
  # http://www.itl.nist.gov/iad/mig//tools/DETware_v2.1.targz.htm
  # cd ~/DETware_v2.1/
  # matlab
  # and run at the matlab prompt:
  # >> scores6
  # >> [P_miss,P_fa] = Compute_DET(target, nontarget)
  # >> Plot_DET(P_miss, P_fa, 'r')
  # that particular result can be compared with the short2-short3 results here
  # http://www.itl.nist.gov/iad/mig/tests/sre/2008/official_results/
  # particularly the telephone-train, telephone-test condition which is here:
  # http://www.itl.nist.gov/iad/mig/tests/sre/2008/official_results/dets/short2-short3.allPrimarySytems.16.det.png
  # Note: this is the only condition we seem to be doing comparable to the systems there,
  # presumably because it matches the training data we used.