[Subject Prev][Subject Next][Thread Prev][Thread Next][Date Index][Thread Index]

[hts-users:01163] 回复: 回复: Re: the speed of making unseen models


Ok,i will use bigger full_all.list .
 
by the way, for this current full_all.list ,is that result gprof gave me correct?

Umud <umud@xxxxxxxx> 写道:
Heiga,you are right.it shouldn't be short.
i used makefile to compile under cygwin. there is the part of my makefile
CC = gcc
CPPFLAGS =
CFLAGS = $(CPPFLAGS) -ansi -D_SVID_SOURCE -DOSS_AUDIO -D'ARCH="i686"' -Wall -Wno-switch -g -O2 -pg 
LDFLAGS = -L/usr/X11R6/lib
LIBS = -lm
INSTALL = /usr/bin/install -c
 
i  added -pg  and -p options and deleted the -O2 options in CFLAGS, also deleted the -O2 options in makefile in HTKTools directory. but gprof gave me the same result. and my HHEd commands are as below
 
./HHEd.exe -A -B -C /cygdrive/e/hts/configs/t
rn.cnf -D -T 1 -p -i -H /cygdrive/e/hts/models/cmp/re_clustered.mmf -w /cygdriv
e/e/hts/models/cmp/re_clustered_all.mmf.1mix /cygdrive/e/hts/edfiles/cmp/mku.he
d /cygdrive/e/hts/data/lists/full.list
 
gprof HHEd.exe gmon.out > result.txt
 
PS: there is only one file in the data/gen directory called 001.lab, and full_all.list file is the same with 001.lab.
 
which part did i wrong?sorry for answered the simple question like this,thank you!
 
 
 


"Heiga ZEN (Byung Ha CHUN)" <zen@xxxxxxxxxxxxxxx> 写道:
Hi,

Umud wrote (2008/02/08 18:40):

> below is the results of HHEd running time via gprof
>
> Each sample counts as 0.01 seconds.
> % cumulative self self total
> time seconds seconds calls ms/call ms/call name
> 14.29 0.65 0.65 Untouch
> 11.43 1.17 0.52 GetCh
> 9.01 1.58 0.41 UntouchV
> 8.79 1.98 0.40 ReadString
> 4.84 2.20 0.22 GetLabId
> 4.62 2.41 0.21 ClearSeenFlags
> 4.40 2.61 0.20 GoNextHMM
> 4.18 2.80 0.19 FindMacroName
> 3.96 2.98 0.18 ClearStreams
> 3.52 3.14 0.16 GetToken
> 2.86 3.27 0.13 GoNextState
> 2.86 3.40 0.13 Hash
> 2.20 3.50 0.10 GetStream
> 1.98 3.59 0.09 fgetc
> 1.76 3.67 0.08 5497 0.01 0.01 GetNode
> 1.76 3.75 0.08 GoNextStream
> 1.76 3.83 0.08 LoadHMMSet
> ...
> 0.00 4.55 0.00 1 0.00 10.00 AddUnseenCommand
> 0.00 4.55 0.00 1 0.00 0.00 ChkedInt
> 0.00 4.55 0.00 1 0.00 170.00 DoEdit
> 0.00 4.55 0.00 1 0.00 0.00 Initialise
> 0.00 4.55 0.00 1 0.00 0.00 SetConfParms
> 0.00 4.55 0.00 1 0.00 0.00 SetTraceCommand
> 0.00 4.55 0.00 1 0.00 0.00 TreeFilter
>
> the total time is 4.55,right?is that time different under windows from linux?

I think this is invalid because the numbers of calls, self ms/call, and total ms/call of many functions are empty.
And 4.55 is too short, your problem is that making unseen models consumed too much time, isn't it?.

How did you compile your HTS?
If you're using gcc, please specify -g and -p, and do not specify optimization options.

Regards,

Heiga ZEN (Byung Ha CHUN)

--
------------------------------------------------
Heiga ZEN (in Japanese pronunciation)
Byung Ha CHUN (in Korean pronunciation)

Department of Computer Science and Engineering
Nagoya Institute of Technology
Gokiso-cho, Showa-ku, Nagoya 466-8555 Japan

http://www.sp.nitech.ac.jp/~zen
------------------------------------------------



雅虎邮箱传递新年祝福,个性贺卡送亲朋!


雅虎邮箱传递新年祝福,个性贺卡送亲朋!
Follow-Ups
[hts-users:01164] RE: [hts-users:01163] 回复: 回复: Re: the speed of making unseen models, Paco Pinto
References
[hts-users:01161] 回复: Re: the speed of making unseen models, Umud