All,

I have a dictation manager listening to the speech coming through the socket for Julius.  I'm trying to add more words and their phonics to my .voca file.  Running the mkdfa.pl scripts works fine, as shown below.  My problem is that if I add over 39 words I get this warning below.  It seems like this warning causes garbage across the socket in my dictation manager.  I went back to use the jclient.pl that comes with Julius and I'm seeing fail messages come across the socket.

How do I fix this?  I don't want to limit the vocabulary.  I'm using acoustics that I have recorded.  I have read on other forums that the acoustics might not good enough this could be the problem.  Can anyone confirm this?

Regards,

Mike

$ perl ../mkdfa.pl test3
test3.grammar has 3 rules
test3.voca    has 3 categories and 42 words
---
Now parsing grammar file
Now modifying grammar to minimize states[-1]
Now parsing vocabulary file
Now making nondeterministic finite automaton[5/5]
Now making deterministic finite automaton[4/6]
* 2 released FA nodes are left on isolated loop
Now making triplet list[4/4]
3 categories, 4 nodes, 4 arcs
-> minimized: 4 nodes, 4 arcs
---
generated: test3.dfa test3.term test3.dict

Julius Error Msg:

------
### read waveform input
pass1_best: <s> KENNETH
pass1_best_wordseq: 0 2
pass1_best_phonemeseq: sil | k eh n ih th
pass1_best_score: -7833.439453
### Recognition: 2nd pass (RL heuristic best-first)
WARNING: 00 _default: hypothesis stack exhausted, terminate search now
STAT: 00 _default: 0 sentences have been found
WARNING: 00 _default: got no candidates, search failed
STAT: 00 _default: 0 generated, 0 pushed, 0 nodes popped in 275
<search failed>
STAT: skip CMN parameter update since last input was invalid

<<< please speak >>>

jclient.pl Listener Ouput:

<INPUT STATUS="LISTEN" TIME="1288828481"/>
.
<INPUT STATUS="STARTREC" TIME="1288828485"/>
.
<STARTRECOG/>
.
<INPUT STATUS="ENDREC" TIME="1288828486"/>
.
<ENDRECOG/>
.
<INPUTPARAM FRAMES="300" MSEC="998"/>
.
<RECOGFAIL/>

--- (Edited on 11/3/2010 7:43 pm [GMT-0500] by gothrog) ---

>WARNING: 00 _default: hypothesis stack exhausted, terminate search now

using Google I get: hypothesis stack exhausted ?? on the Julius Sourceforge forum... the answer seems to be:

I worked around it by increasing the threshold for maximum number of expanded hypothesis
-m 5000

Please let us know if this fixes your problem too,

thanks

--- (Edited on 11/4/2010 3:33 pm [GMT-0400] by kmaclean) ---

I have tried that already..... and it has not worked for me.  This is why I'm wondering if my acoustics are not good enough.  I saved them as waves of 48000hz and 16 Bit.

--- (Edited on 11/4/2010 2:55 pm [GMT-0500] by gothrog) ---

Ken,

I seem to have fixed this one problem.  I changed the "-b" option in the search parameters.  Are you aware of potential side effects that may occur from doing this?

Mike

######################################################################
#### Search Parameters
######################################################################
-b 0
#-b 400                 # beam width on 1st pass (#nodes) for monophone
#-b 800                 # beam width on 1st pass (#nodes) for triphone,PTM
#-b 1000                # beam width on 1st pass (#nodes) for triphone,PTM,engine=v2.1
-b2 200                 # beam width on 2nd pass (#words)
-sb 200.0        # score beam envelope threshold
#-s 500                 # hypotheses stack size on 2nd pass (#hypo)
##########-s 1000                 # hypotheses stack size on 2nd pass (#hypo)
#-m 5000                # hypotheses overflow threshold (#hypo)
#-m 2000                # hypotheses overflow threshold (#hypo)
#-lookuprange 5         # lookup range for word expansion (#frame)
#-n 1                   # num of sentences to find (#sentence)
#-n 10                  #   (default for 'standard' configuration)
#-output 1              # num of found sentences to output (#sentence)
#-looktrellis        # search within only backtrellis word

--- (Edited on 11/4/2010 3:46 pm [GMT-0500] by gothrog) ---