VoxForge
VoxForge
Dear all
I want to test recognizer with training file (using label).
But I 'm confuse about HResults's use parameter.
I try with HResults -f referenceFile triphonelistFile recogOutputFile
But output are always show 100% correct, why?
And when I swap parameter, it show
ERROR [+6570] GetLabelList: n[1] > numLists[0]
please tell me how to use it.
Best.
ps.
triphoneList example
#!MLF!#
"*/sx1.lab"
sil
n+aa
n-aa+j^
aa-j^+s
j^-s+a
s-a+ng
a-ng+aa
ng-aa+s
aa-s+a
s-a+p^
a-p^+sp
p^-sp+ph
sp-ph+a
ph-a+s
a-s+ii
s-ii
sil
.
"*/sx2.lab"
sil
d+@
r
reference example
#!MLF!#
"*/sx1.lab"
sil
naaj^
sangaa
sap^phasii
sil
.
"*/sx2.lab"
sil
d@k^tqq
......
output file
#!MLF!#
"*/sx1.rec"
2000000 4900000 mxx -1235.608887
4900000 6400000 khrx -592.565735
6400000 9600000 ngaaj^ -1016.047363
9600000 16300000 sap^phasii -672.542603
16300000 18900000 ziik^ -121.680260
.
"*/sx2.rec"
2100000 5000000 d@k^tqq -377.382629
5000000 8700000 caaj^ -1005.422913
--- (Edited on 3/30/2011 3:12 am [GMT-0500] by Noah) ---
--- (Edited on 3/30/2011 3:35 am [GMT-0500] by Noah) ---
>But I 'm confuse about HResults's use parameter.
see here: Test Acoustic Model Using HTK
--- (Edited on 3/30/2011 10:24 am [GMT-0400] by kmaclean) ---
Thanks.
Now I understand what the problem.
I can get recognizer's performance result when I test on recognizer's output which has less output sentences(files) when compare to training sentences(files) (It work when test on 1 sentence). (1 test sentence : 1680 training sentences)
But it doesn't work when I try to test with all training file. I want result of all traning files to see performance.
It return error when I test with all training files
ERROR [+6570] GetLabelList: n[1] > numLists[0]
I try by config test sentences less than training sentences (1679:1680) but it's doesn't work too.
How to solve this problem?
Best.
--- (Edited on 3/31/2011 2:59 am [GMT-0500] by Noah) ---
>ERROR [+6570] GetLabelList: n[1] > numLists[0]
what does the HTK book say about this error? Have you checked the HTK email archives?
--- (Edited on 4/11/2011 9:30 am [GMT-0400] by kmaclean) ---