sandbox/language-id: Updating logistic regression conf files for mixture model. Also updating run_logistic_regression.sh for new results.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/language_id@3816 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

egs/lre/v1/conf/logistic-regression.conf

@@ -1,3 +1,5 @@
 --max-steps=20
---normalizer=0.00325
---verbose=2
+--normalizer=0.00025
+--verbose=3
+--power=0.15
+--mix-up=90

egs/lre/v1/run.sh

@@ -36,7 +36,7 @@ src_list="data/sre08_train_10sec_female \
 
 # Remove any spk2gender files that we have: since not all data
 # sources have this info, it will cause problems with combine_data.sh
-for d in $src_list; do rm $d/spk2gender 2>/dev/null; done
+for d in $src_list; do rm -f $d/spk2gender 2>/dev/null; done
 
 utils/combine_data.sh data/train_unsplit $src_list
 

egs/lre/v1/run_logistic_regression.sh

@@ -66,8 +66,8 @@ cat exp/ivectors_train/posteriors | \
 compute-wer --mode=present --text ark:<(utils/remove_dialect.pl data/train/utt2lang) \
   ark:exp/ivectors_train/output
 
-# %WER 5.35 [ 3834 / 71668, 0 ins, 0 del, 3834 sub ] [PARTIAL]
-# %SER 5.35 [ 3834 / 71668 ]
+# %WER 4.73 [ 3389 / 71668, 0 ins, 0 del, 3389 sub ] [PARTIAL]
+# %SER 4.73 [ 3389 / 71668 ]
 # Scored 71668 sentences, 16 not present in hyp.
 
 #  scp:exp/ivectors_lre07/ivector.scp ark,t:- | \
@@ -84,6 +84,6 @@ logistic-regression-eval $model_rebalanced \
 compute-wer --text ark:<(utils/remove_dialect.pl data/lre07/utt2lang) \
   ark:exp/ivectors_lre07/output
 > compute-wer --text ark:/dev/fd/63 ark:exp/ivectors_lre07/output 
-# %WER 33.84 [ 2547 / 7527, 0 ins, 0 del, 2547 sub ]
-# %SER 33.84 [ 2547 / 7527 ]
+# %WER 33.04 [ 2487 / 7527, 0 ins, 0 del, 2487 sub ]
+# %SER 33.04 [ 2487 / 7527 ]
 # Scored 7527 sentences, 0 not present in hyp.