How do I get a silent phoneme in the second path?

[Hiroshiba]

I'm trying speech recognition using an automaton, but silent phonemes (SP) are never recognized. SP phonemes appear in the first path, but not in the second path.

I tried with these commands and parameters:

# hoge.grammar
S       :  NS_B FRUIT_N NOISE PLEASE NS_E
FRUIT_N :  FRUIT
PLEASE  :  DESU

# hoge.voca
% FRUIT
蜜柑           m i k a N
リンゴ         r i N g o
ぶどう         b u d o:
% DESU
です           d e s u
% NS_B
<s>            silB
% NS_E
</s>           silE
% NOISE
<sp>           sp

# wave file
curl https://raw.githubusercontent.com/Hiroshiba/julius/26c00a15f8391f3d7ef804203903223d2021a65f/input-16000.wav > input-16000.wav

# command
mkdfa.pl hoge

echo input-16000.wav | \
julius \
  -input file \
  -h segmentation-kit/models/hmmdefs_monof_mix16_gid.binhmm \
  -gram hoge \
  -n 10 \

and got this result:

...

### Recognition: 1st pass (LR beam)
pass1_best: <s> リンゴ <sp> です </s>
pass1_best_wordseq: 2 0 4 1 3
pass1_best_phonemeseq: silB | r i N g o | sp | d e s u | silE
pass1_best_score: -5970.041504
### Recognition: 2nd pass (RL heuristic best-first)
WARNING: 00 _default: hypothesis stack exhausted, terminate search now
STAT: 00 _default: 2 sentences have been found
STAT: 00 _default: 8 generated, 8 pushed, 10 nodes popped in 268
sentence1: <s> リンゴ です </s>
wseq1: 2 0 1 3
phseq1: silB | r i N g o | d e s u | silE
cmscore1: 1.000 0.973 1.000 1.000
score1: -5970.035156

...

and these are dfa file and dict file:

# hoge.dfa
0 3 1 0 0
1 1 2 0 0
2 4 3 0 0
3 0 4 0 0
4 2 5 0 0
5 -1 -1 1 0

# hoge.dict
0       [蜜柑]  m i k a N
0       [リンゴ]        r i N g o
0       [ぶどう]        b u d o:
1       [です]  d e s u
2       [<s>]   silB
3       [</s>]  silE
4       [<sp>]  sp

I thought SP is easier to appear with the positive value -transp, but SP was not appeared.

How do I get a silent phoneme in the second path?