VoxForge
Hi,
I am new at sphinx and I have been experiencing some trouble to install it.
I am using Linux version 2.6.18-4-686 (Debian 2.6.18.dfsg.1-12), gcc 4.1.2, perl 5.8.8 and the nightly.tar.gz versions of an4, Sphinxtrain, sphinxbase and sphinx3.
I have been following the tutorial of http://www.speech.cs.cmu.edu/sphinx/tutorial.html page.
Everything is okay until I try to perform a preliminary decode using the following command:
perl scripts_pl/decode/slave.pl
Does anyone knows how to solve the following error:
MODULE: DECODE Decoding using models previously trained
Decoding 130 segments starting at 0 (part 1 of 1)
0%
Can't open /home/margarita/tutorial/an4/result/an4-1-1.match
SENTENCE ERROR: 100.000% (130/130)
I looked up into the an4/result directory and I found an4.align and an4.match
What can I do?
Thanks
--- (Edited on 4/1/2008 6:06 pm [GMT-0500] by m811) ---
Hi,
Thanks for answering. The files are identical. I tried to compile everything again and after I do
cd SphinxTrain
./configure
make
perl scripts/setup_tutorial.pl an4
cd ../an4
perl scripts_pl/make_feats.pl -ctl etc/an4_train.fileids
perl scripts_pl/RunAll.pl
I got the following errors:
At module 30.cd_hmm_untied Training Context Dependent models for sub-word units
(triphones) with untied states
ERROR: "backward.c", line 401: final state not reached
ERROR: "baum_welch.c", line 331: mjdr/cen2-mjdr-b ignored
At module 50.cd-hmm_tied Training Context dependent models for intermediate CD-tied
models.
ERROR: "backward.c", line 401: final state not reached
ERROR: "baum_welch.c", line 331: fash/cen7-fash-b ignored
ERROR: "backward.c", line 401: final state not reached
ERROR: "baum_welch.c", line 331: meab/cen2-meab-b ignored
ERROR: "gauden.c", line 1700: var (mgau= 708, feat= 0, density=0, component=0) < 0
FATAL_ERROR: "s3gau_io.c", line 364: Checksum error; read corrupt data.
Failed to start norm
FATAL_ERROR: "s3gau_io.c", line 364: Checksum error; read corrupt data.
What could you suggest me to do?
Thanks
--- (Edited on 4/2/2008 12:32 pm [GMT-0500] by m811) ---
> ERROR: "backward.c", line 401: final state not reached
> ERROR: "baum_welch.c", line 331: fash/cen7-fash-b ignored
> ERROR: "backward.c", line 401: final state not reached
> ERROR: "baum_welch.c", line 331: meab/cen2-meab-b ignored
> ERROR: "gauden.c", line 1700: var (mgau= 708, feat= 0, density=0, component=0) < 0
Thouse are not critical, it's caused by bad transcription and too big number of senones. Ignore them.
> FATAL_ERROR: "s3gau_io.c", line 364: Checksum error; read corrupt data.
Failed to start norm
Are you running on some strange architecture, on 64 bit probably? If the data is corrupted there might be many reasons for that. Probably training was stopped somewhere in the middle or something else.
I've just tried it and everything works fine for me. So please try again from the scratch. Recompile sphinxtrain, unpack an4 and start training. If it will still fail, upload all files from an4 without audio data or only logs somewhere and give us a link.
--- (Edited on 4/3/2008 1:17 am [GMT-0500] by nsh) ---
Hi,
Thanks for answering again.
I redo all the process and I do not get the FATAL_ERROR anymore but now, when I try to perform a preliminary decode I get these
debian:/home/margarita/tutorial/an4# perl scripts_pl/decode/slave.pl MODULE: DECODE Decoding using models previously trained
Decoding 130 segments starting at 0 (part 1 of 1)
0%
This step had 6 ERROR messages and 2 WARNING messages. Please check the log file for details.
SENTENCE ERROR: 56.923% (74/130)
As you told me before, I tried to copy the sphinx3_decode from
/home/margarita/tutorial/sphinx3/build/bin/sphinx3_decode
into
/home/margarita/tutorial/an4/bin/sphinx3_decode
I do not have a web page but ,I created a new email account where I loaded my logdir directory if it is okay with you here is the information:
password:ma28v9t0
Thanks for everything
--- (Edited on 4/3/2008 6:04 pm [GMT-0500] by m811) ---
> SENTENCE ERROR: 56.923% (74/130)
It's the expected result. You finished tutorial successfully.
Heh, not a straightforward way. Next time you can use public services like http://mediafire.com to upload files.
--- (Edited on 4/3/2008 6:15 pm [GMT-0500] by nsh) ---
Hi,
I was follwoing Robust Group Tutorial. I used pocketsphinx. All seem worked, but at the end I get the following:
an4# perl scripts_pl/decode/slave.pl
MODULE: DECODE Decoding using models previously trained
Decoding 130 segments starting at 0 (part 1 of 1)
0%
WARNING: This step had 0 ERROR messages and 23 WARNING messages. Please check the log file for details.
SENTENCE ERROR: 100.000% (130/130)
Please help.
--- (Edited on 6/25/2008 3:43 pm [GMT-0500] by Visitor) ---
Log file contains no ERROR. But few warnings. The log file can be accessed here:
http://sip.hq.com.au/an4-1-1.log
Best Regards,
Zia
--- (Edited on 6/25/2008 8:35 pm [GMT-0500] by Visitor) ---