Menu

Acoustic Model Adaptation - problem with BW

Help
2011-09-18
2012-09-22
  • Dhiraj Gupta

    Dhiraj Gupta - 2011-09-18

    Hi,

    There seems to be some problem and my posting on Sphinx-4 are not getting
    saved so I am putting it here.

    I am trying to adapt the acoustic model. I could run sphinx_fe but when I try
    to run BW.exe, it doesn;t seem to run fine. I have gone through the tutorial
    and the previous related threads but still am not able to find the problem.

    I have placed the files at http://dl.dropbox.com/u/39473965/BW_Data.zip

    BW.exe stops working after the giving the below output. Can you please suggest
    what the problem could be?

    INFO: main.c(197): Compiled on Apr 16 2011 at 02:55:53
    INFO: cmd_ln.c(559): Parsing command line:
    bw \
    -hmmdir wsj_8kHz \
    -moddeffn wsj_8kHz\mdef.txt \
    -ts2cbfn .semi. \
    -feat 1s_c_d_dd \
    -svspec 0-12/13-25/26-38 \
    -cmn current \
    -agc none \
    -dictfn arctic20.dic \
    -ctlfn arctic20.fileids \
    -lsnfn arctic20.transcription \
    -accumdir .

    Current configuration:

    -2passvar no no
    -abeam 1e-100 1.000000e-100
    -accumdir .
    -agc none none
    -agcthresh 2.0 2.000000e+000
    -bbeam 1e-100 1.000000e-100
    -cb2mllrfn .1cls. .1cls.
    -cepdir
    -cepext .mfc .mfc
    -ceplen 13 13
    -cepwin 0 0
    -ckptintv 0
    -cmn current current
    -cmninit 8.0 8.0
    -ctlfn arctic20.fileids
    -diagfull no no
    -dictfn arctic20.dic
    -example no no
    -fdictfn
    -feat 1s_c_d_dd 1s_c_d_dd
    -fullsuffixmatch no no
    -fullvar no no
    -help no no
    -hmmdir wsj_8kHz
    -latdir
    -latext
    -lda
    -ldaaccum no no
    -ldadim 0 0
    -lsnfn arctic20.transcription
    -ltsoov no no
    -lw 11.5 1.150000e+001
    -maxuttlen 0 0
    -meanfn
    -meanreest yes yes
    -mixwfn
    -mixwreest yes yes
    -mllrmat
    -mmie no no
    -mmie_type rand rand
    -moddeffn wsj_8kHz\mdef.txt
    -mwfloor 0.00001 1.000000e-005
    -npart 0
    -nskip 0
    -outphsegdir
    -outputfullpath no no
    -part 0
    -pdumpdir
    -phsegdir
    -phsegext phseg phseg
    -runlen -1 -1
    -sentdir
    -sentext sent sent
    -silcomp none none
    -siltag SIL SIL
    -spthresh 0.0 0.000000e+000
    -svspec 0-12/13-25/26-38
    -timing yes yes
    -tmatfn
    -tmatreest yes yes
    -topn 4 4
    -tpfloor 0.0001 1.000000e-004
    -ts2cbfn .semi.
    -varfloor 0.00001 1.000000e-005
    -varfn
    -varnorm no no
    -varreest yes yes
    -viterbi no no

    INFO: feat.c(697): Initializing feature stream to type: '1s_c_d_dd',
    ceplen=13,
    CMN='current', VARNORM='no', AGC='none'
    INFO: cmn.c(142): mean= 12.00, mean= 0.0
    INFO: main.c(222): Using subvector specification 0-12/13-25/26-38
    INFO: main.c(286): Reading wsj_8kHz\mdef.txt
    INFO: model_def_io.c(587): Model definition info:
    INFO: model_def_io.c(588): 110883 total models defined (54 base, 110829 tri)
    INFO: model_def_io.c(589): 443532 total states
    INFO: model_def_io.c(590): 4162 total tied states
    INFO: model_def_io.c(591): 162 total tied CI states
    INFO: model_def_io.c(592): 54 total tied transition matrices
    INFO: model_def_io.c(593): 4 max state/model
    INFO: model_def_io.c(594): 4 min state/model

    Thanks & Regards,
    Dhiraj

     
  • Nickolay V. Shmyrev

    There seems to be some problem and my posting on Sphinx-4 are not getting
    saved so I am putting it here.

    Sphinx-4 is only for the discussion about Sphinx 4. This question is not
    appropriate there

    I have gone through the tutorial and the previous related threads but still
    am not able to find the problem.

    The bw application is crashed. You need provide a stacktrace in order to
    diagnose the problem. You also need to mention the version of the Sphinxtrain
    you are using.

    http://fedoraproject.org/wiki/StackTraces

     
  • Dhiraj Gupta

    Dhiraj Gupta - 2011-09-19

    Hi,

    Thanks for information. I installed sphinxbase and sphinxtrain 0.7 from the
    below link

    http://cmusphinx.sourceforge.net/wiki/download

    Also, I installed GDB and below is the result. There were addidtional lines
    about ntdll.dll at the end. Please let me know if I need to provide any
    further information.

    Starting program: C:\sphinxtrain-1.0.7-win32/bw.exe -hmmdir wsj_8kHz -moddeffn
    w
    sj_8kHz\mdef.txt -ts2cbfn .semi. -feat 1s_c_d_dd -svspec 0-12/13-25/26-38 -cmn
    c
    urrent -agc none -dictfn arctic20.dic -ctlfn arctic20.fileids -lsnfn
    arctic20.tr
    anscription -accumdir .

    INFO: main.c(197): Compiled on Apr 16 2011 at 02:55:53
    INFO: cmd_ln.c(559): Parsing command line:
    C:\sphinxtrain-1.0.7-win32/bw.exe \
    -hmmdir wsj_8kHz \
    -moddeffn wsj_8kHz\mdef.txt \
    -ts2cbfn .semi. \
    -feat 1s_c_d_dd \
    -svspec 0-12/13-25/26-38 \
    -cmn current \
    -agc none \
    -dictfn arctic20.dic \
    -ctlfn arctic20.fileids \
    -lsnfn arctic20.transcription \
    -accumdir .

    Current configuration:

    -2passvar no no
    -abeam 1e-100 1.000000e-100
    -accumdir .
    -agc none none
    -agcthresh 2.0 2.000000e+000
    -bbeam 1e-100 1.000000e-100
    -cb2mllrfn .1cls. .1cls.
    -cepdir
    -cepext .mfc .mfc
    -ceplen 13 13
    -cepwin 0 0
    -ckptintv 0
    -cmn current current
    -cmninit 8.0 8.0
    -ctlfn arctic20.fileids
    -diagfull no no
    -dictfn arctic20.dic
    -example no no
    -fdictfn
    -feat 1s_c_d_dd 1s_c_d_dd
    -fullsuffixmatch no no
    -fullvar no no
    -help no no
    -hmmdir wsj_8kHz
    -latdir
    -latext
    -lda
    -ldaaccum no no
    -ldadim 0 0
    -lsnfn arctic20.transcription
    -ltsoov no no
    -lw 11.5 1.150000e+001
    -maxuttlen 0 0
    -meanfn
    -meanreest yes yes
    -mixwfn
    -mixwreest yes yes
    -mllrmat
    -mmie no no
    -mmie_type rand rand
    -moddeffn wsj_8kHz\mdef.txt
    -mwfloor 0.00001 1.000000e-005
    -npart 0
    -nskip 0
    -outphsegdir
    -outputfullpath no no
    -part 0
    -pdumpdir
    -phsegdir
    -phsegext phseg phseg
    -runlen -1 -1
    -sentdir
    -sentext sent sent
    -silcomp none none
    -siltag SIL SIL
    -spthresh 0.0 0.000000e+000
    -svspec 0-12/13-25/26-38
    -timing yes yes
    -tmatfn
    -tmatreest yes yes
    -topn 4 4
    -tpfloor 0.0001 1.000000e-004
    -ts2cbfn .semi.
    -varfloor 0.00001 1.000000e-005
    -varfn
    -varnorm no no
    -varreest yes yes
    -viterbi no no

    INFO: feat.c(697): Initializing feature stream to type: '1s_c_d_dd',
    ceplen=13,
    CMN='current', VARNORM='no', AGC='none'
    INFO: cmn.c(142): mean= 12.00, mean= 0.0
    INFO: main.c(222): Using subvector specification 0-12/13-25/26-38
    INFO: main.c(286): Reading wsj_8kHz\mdef.txt
    INFO: model_def_io.c(587): Model definition info:
    INFO: model_def_io.c(588): 110883 total models defined (54 base, 110829 tri)
    INFO: model_def_io.c(589): 443532 total states
    INFO: model_def_io.c(590): 4162 total tied states
    INFO: model_def_io.c(591): 162 total tied CI states
    INFO: model_def_io.c(592): 54 total tied transition matrices
    INFO: model_def_io.c(593): 4 max state/model
    INFO: model_def_io.c(594): 4 min state/model

    Program received signal SIGSEGV, Segmentation fault.
    0x7774fa7b in ntdll!memchr () from C:\Windows\system32\ntdll.dll
    (gdb)

    Thanks & Regards,
    Dhiraj

     
  • Nickolay V. Shmyrev

    I installed sphinxbase and sphinxtrain 0.7 from the below link

    Please install snapshot

    Segmentation fault. 0x7774fa7b in ntdll!memchr () from
    C:\Windows\system32\ntdll.dll (gdb)

    You need to type "bt" command now to get a stack trace. Please check the link
    above for more information how backtrace looks.

     
  • Dhiraj Gupta

    Dhiraj Gupta - 2011-09-19

    Thanks for quick reply. I will try with snapshot version too.

    Also, I wrote "bt" and below is the trace I got. One more point I forgot to
    mention, I installed msvcr100.dll from the link http://www.dll-
    files.com/dllindex/dll-files.shtml?msvcr100
    as initially I got message about this missing DLL.

    (gdb) bt

    0 0x7774fa7b in ntdll!memchr () from C:\Windows\system32\ntdll.dll

    1 0x7774f98a in ntdll!memchr () from C:\Windows\system32\ntdll.dll

    2 0x6a892ee7 in lock_file () from C:\sphinxtrain-1.0.7-win32\msvcr100.dll

    3 0x6a894280 in fread_s () from C:\sphinxtrain-1.0.7-win32\msvcr100.dll

    4 0x00fda5a8 in ?? ()

    5 0x6a8942d6 in fread () from C:\sphinxtrain-1.0.7-win32\msvcr100.dll

    6 0x100200a9 in sphinxbase!fread_retry ()

    from C:\sphinxtrain-1.0.7-win32\sphinxbase.dll

    7 0x00fa3af5 in ?? ()

    8 0x00fa2d53 in ?? ()

    9 0x00fa3014 in ?? ()

    10 0x00fa0a31 in ?? ()

    11 0x00fa5e80 in ?? ()

    12 0x00f95b6a in ?? ()

    13 0x00f99401 in ?? ()

    14 0x00fa8150 in ?? ()

    15 0x76301114 in KERNEL32!AcquireSRWLockExclusive ()

    from C:\Windows\system32\kernel32.dll

    16 0x7777b429 in ntdll!RtlInsertElementGenericTableAvl ()

    from C:\Windows\system32\ntdll.dll

    17 0x7777b3fc in ntdll!RtlInsertElementGenericTableAvl ()

    from C:\Windows\system32\ntdll.dll

    18 0x00000000 in ?? ()

    Thanks & REgards,
    Dhiraj

     
  • Dhiraj Gupta

    Dhiraj Gupta - 2011-09-19

    Hi,

    I had been using the BW.exe available in sphinxtrain-1.0.7-win32.zip so I
    haven't compiled the code till now. For snapshot, I see it just has the source
    code. I know it might be very basic question but is there way to compile it? I
    mean like for Sphinx-4, we have ANT utility that does the work, is there
    something similar for creating BW.exe too?

    Thanks & Regards,
    Dhiraj

     
  • Dhiraj Gupta

    Dhiraj Gupta - 2011-09-19

    I tried to search "fread_s msvcr100.dll" and found some information at the
    link http://cmusphinx.sourceforge.net/wiki/faq Under the Q. Pocketsphinx crashes on Windows in _lock_file, it
    says as below. But I couldn;t make much out of it. Please help!

    sphinxbase was compiled iwth MultiThreadedDLL runtime, see in vcxproj

    <runtimelibrary>MultiThreadedDLL</runtimelibrary>

    If you don't compile your project with similar setting it will crash. Use
    proper runtime library or recompile sphinxbase

    Thanks & Regards,
    Dhiraj

     
  • Nickolay V. Shmyrev

    Just download the snapshot and recompile it using Visual Studio 2010. See
    README file for details. This bug was already fixed.

     
  • Dhiraj Gupta

    Dhiraj Gupta - 2011-09-21

    Thanks much for the information.

    Thanks & Regards,
    Dhiraj

     
  • Dhiraj Gupta

    Dhiraj Gupta - 2011-10-03

    Hi,

    Thanks for the help and I could run the BW.exe. But when I run it, both the
    utterances are getting ignored. Please find below the dump. I could find some
    related info on http://www.speech.cs.cmu.edu/sphinxman/fr6.html and http://cmusphinx.sourceforge.net/wik
    i/tutorialam
    but still am
    not clear what should be done so that the error goes off.

    utt> 0 arctic_0001 1023INFO: cmn.c(175): CMN: 9.20 -0.28 0.1
    3 -0.17 -0.30 -0.27 -0.13 -0.17 -0.08 -0.13 -0.11 -0.13 -0.10
    0 528 187 ERROR: "backward.c", line 430: Failed to align audio to trancrip
    t: final state of the search is not reached
    ERROR: "baum_welch.c", line 333: arctic_0001 ignored
    utt 0.012x 1.122e upd 0.012x 0.994e fwd 0.012x 0.994e bwd 0.000x 0.000e gau
    0.0
    02x 0.962e rsts 0.000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 1 arctic_0002 117INFO: cmn.c(175): CMN: 7.29 -0.12 -0.0
    3 -0.17 -0.33 -0.21 -0.14 -0.10 -0.05 -0.08 -0.04 -0.08 -0.07
    0 92 65 WARNING: "gauden.c", line 1343: Scaling factor too small: -92912.
    335564
    ERROR: "backward.c", line 1019: alpha(3.683982e-043) <> sum of alphas * betas
    (0
    .000000e+000) in frame 115
    ERROR: "baum_welch.c", line 333: arctic_0002 ignored
    utt 0.000x 0.000e upd 0.000x 0.000e fwd 0.000x 0.000e bwd 0.000x 0.000e gau
    0.0
    00x 0.000e rsts 0.000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    overall> WIN32(N/A) 0 (-0) 0.000000e+000 0.000000e+000 0.000x 1.250e
    WARNING: "accum.c", line 618: Over 500 senones never occur in the input data.
    Th
    is is normal for context-dependent untied senone training or for adaptation,
    but
    could indicate a serious problem otherwise.

    Thanks & Regards,
    Dhiraj

     
  • Nickolay V. Shmyrev

    If alignment mismatch consistently it means you extracted mfc features
    incorrectly. For example your files were stereo or had incorrect sample rate.
    You need to provide the files in order to let us reproduce your trouble if you
    need more help.

     
  • Dhiraj Gupta

    Dhiraj Gupta - 2011-11-25

    Hi,

    I got involved in something else and was not able to try it again. Got to try
    again only today after a month's time. Continuing the same thread. I have put
    the files at below link. The message is alos put below which says the files
    are being ignored. Please let me know if there is more information required.
    Can you please check and let me know what needs to be done to make it work?

    http://dl.dropbox.com/u/39473965/BW%20Files.zip

    INFO: main.c(194): Compiled on Nov 24 2011 at 14:09:23
    INFO: cmd_ln.c(691): Parsing command line:
    bw \
    -hmmdir wsj_8kHz \
    -moddeffn wsj_8kHz\mdef.txt \
    -ts2cbfn .semi. \
    -feat 1s_c_d_dd \
    -cmn current \
    -agc none \
    -dictfn arctic20.dic \
    -ctlfn arctic20.listoffiles \
    -lsnfn arctic20.transcription \
    -accumdir .

    Current configuration:

    -2passvar no no
    -abeam 1e-100 1.000000e-100
    -accumdir .
    -agc none none
    -agcthresh 2.0 2.000000e+000
    -bbeam 1e-100 1.000000e-100
    -cb2mllrfn .1cls. .1cls.
    -cepdir
    -cepext mfc mfc
    -ceplen 13 13
    -ckptintv 0
    -cmn current current
    -cmninit 8.0 8.0
    -ctlfn arctic20.listoffiles
    -diagfull no no
    -dictfn arctic20.dic
    -example no no
    -fdictfn
    -feat 1s_c_d_dd 1s_c_d_dd
    -fullsuffixmatch no no
    -fullvar no no
    -help no no
    -hmmdir wsj_8kHz
    -latdir
    -latext
    -lda
    -ldaaccum no no
    -ldadim 0 0
    -lsnfn arctic20.transcription
    -ltsoov no no
    -lw 11.5 1.150000e+001
    -maxuttlen 0 0
    -meanfn
    -meanreest yes yes
    -mixwfn
    -mixwreest yes yes
    -mllrmat
    -mmie no no
    -mmie_type rand rand
    -moddeffn wsj_8kHz\mdef.txt
    -mwfloor 0.00001 1.000000e-005
    -npart 0
    -nskip 0
    -outphsegdir
    -outputfullpath no no
    -part 0
    -pdumpdir
    -phsegdir
    -phsegext phseg phseg
    -runlen -1 -1
    -sentdir
    -sentext sent sent
    -spthresh 0.0 0.000000e+000
    -svspec
    -timing yes yes
    -tmatfn
    -tmatreest yes yes
    -topn 4 4
    -tpfloor 0.0001 1.000000e-004
    -ts2cbfn .semi.
    -varfloor 0.00001 1.000000e-005
    -varfn
    -varnorm no no
    -varreest yes yes
    -viterbi no no

    INFO: feat.c(684): Initializing feature stream to type: '1s_c_d_dd',
    ceplen=13,
    CMN='current', VARNORM='no', AGC='none'
    INFO: cmn.c(142): mean= 12.00, mean= 0.0
    INFO: main.c(283): Reading wsj_8kHz\mdef.txt
    INFO: model_def_io.c(573): Model definition info:
    INFO: model_def_io.c(574): 110883 total models defined (54 base, 110829 tri)
    INFO: model_def_io.c(575): 443532 total states
    INFO: model_def_io.c(576): 4162 total tied states
    INFO: model_def_io.c(577): 162 total tied CI states
    INFO: model_def_io.c(578): 54 total tied transition matrices
    INFO: model_def_io.c(579): 4 max state/model
    INFO: model_def_io.c(580): 4 min state/model
    INFO: s3mixw_io.c(116): Read wsj_8kHz/mixture_weights
    INFO: s3tmat_io.c(115): Read wsj_8kHz/transition_matrices
    INFO: mod_inv.c(300): inserting tprob floor 1.000000e-004 and renormalizing
    INFO: s3gau_io.c(166): Read wsj_8kHz/means
    INFO: s3gau_io.c(166): Read wsj_8kHz/variances
    INFO: gauden.c(184): 4162 total mgau
    INFO: gauden.c(158): 1 feature streams (|0|=39 )
    INFO: gauden.c(195): 8 total densities
    INFO: gauden.c(97): min_var=1.000000e-005
    INFO: gauden.c(173): compute 4 densities/frame

    of codebooks in mean/var files, 4162, inconsistent with ts2cb mapping 1

    INFO: main.c(395): Will reestimate mixing weights.
    INFO: main.c(397): Will reestimate means.
    INFO: main.c(399): Will reestimate variances.
    INFO: main.c(407): Will reestimate transition matrices
    INFO: main.c(420): Reading main lexicon: arctic20.dic
    INFO: lexicon.c(218): 129246 entries added from arctic20.dic
    INFO: main.c(432): Reading filler lexicon: wsj_8kHz/noisedict
    INFO: lexicon.c(218): 27 entries added from wsj_8kHz/noisedict
    INFO: corpus.c(1078): Will process all remaining utts starting at 0
    INFO: main.c(639): Reestimation: Baum-Welch
    INFO: main.c(644): Generating profiling information consumes significant CPU
    res
    ources.
    INFO: main.c(645): If you are not interested in profiling, use -timing no
    column defns
    <seq>
    <id>
    <n_frame_in>
    <n_frame_del>
    <n_state_shmm>
    <avg_states_alpha>
    <avg_states_beta>
    <avg_states_reest>
    <avg_posterior_prune>
    <frame_log_lik>
    <utt_log_lik>
    ... timing info ...
    utt> 0 arctic_0001 405INFO: cmn.c(175): CMN: 7.57 0.30 -0.1
    9 0.06 -0.32 -0.12 -0.32 -0.20 -0.32 -0.15 -0.13 -0.09 -0.07
    0 40 37 WARNING: "gauden.c", line 1343: Scaling factor too small: -33761.
    377162
    ERROR: "backward.c", line 1019: alpha(1.196170e-018) <> sum of alphas * betas
    (0
    .000000e+000) in frame 403
    ERROR: "baum_welch.c", line 333: arctic_0001 ignored
    utt 0.004x 4.167e upd 0.004x 1.859e fwd 0.004x 1.090e bwd 0.000x 0.000e gau
    0.0
    04x 0.513e rsts 0.000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 1 arctic_0002 442INFO: cmn.c(175): CMN: 7.51 0.30 -0.1
    8 0.15 -0.28 -0.13 -0.28 -0.19 -0.36 -0.12 -0.14 -0.02 -0.08
    0 56 52 WARNING: "gauden.c", line 1343: Scaling factor too small: -33458.
    883087
    ERROR: "backward.c", line 1019: alpha(9.822728e-027) <> sum of alphas * betas
    (0
    .000000e+000) in frame 440
    ERROR: "baum_welch.c", line 333: arctic_0002 ignored
    utt 0.007x 1.538e upd 0.007x 1.122e fwd 0.007x 0.705e bwd 0.000x 0.000e gau
    0.0
    00x 0.000e rsts 0.000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 2 arctic_0003 368INFO: cmn.c(175): CMN: 7.20 0.22 -0.1
    7 0.11 -0.27 -0.09 -0.28 -0.23 -0.31 -0.12 -0.09 -0.04 -0.08
    0 44 41 WARNING: "gauden.c", line 1343: Scaling factor too small: -31655.
    385448
    ERROR: "backward.c", line 1019: alpha(4.485360e-019) <> sum of alphas * betas
    (0
    .000000e+000) in frame 366
    ERROR: "baum_welch.c", line 333: arctic_0003 ignored
    utt 0.000x 0.000e upd 0.000x 0.000e fwd 0.000x 0.000e bwd 0.000x 0.000e gau
    0.0
    00x 0.000e rsts 0.000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 3 arctic_0004 284INFO: cmn.c(175): CMN: 7.19 0.30 -0.1
    2 0.13 -0.20 -0.18 -0.31 -0.22 -0.31 -0.15 -0.13 -0.05 -0.06
    0 32 30 WARNING: "gauden.c", line 1343: Scaling factor too small: -31715.
    724708
    ERROR: "backward.c", line 1019: alpha(5.516553e-013) <> sum of alphas * betas
    (0
    .000000e+000) in frame 282
    ERROR: "baum_welch.c", line 333: arctic_0004 ignored
    utt 0.005x 1.795e upd 0.005x 1.218e fwd 0.005x 0.577e bwd 0.000x 0.000e gau
    0.0
    05x 0.449e rsts 0.000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 4 arctic_0005 340INFO: cmn.c(175): CMN: 7.19 0.34 -0.2
    1 0.09 -0.26 -0.17 -0.28 -0.19 -0.34 -0.12 -0.12 -0.04 -0.07
    0 48 45 WARNING: "gauden.c", line 1343: Scaling factor too small: -31662.
    944220
    ERROR: "backward.c", line 1019: alpha(3.639718e-021) <> sum of alphas * betas
    (0
    .000000e+000) in frame 338
    ERROR: "baum_welch.c", line 333: arctic_0005 ignored
    utt 0.009x 1.090e upd 0.009x 0.737e fwd 0.005x 0.833e bwd 0.000x 0.000e gau
    0.0
    00x 0.000e rsts 0.000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 5 arctic_0006 373INFO: cmn.c(175): CMN: 7.73 0.55 -0.1
    7 0.01 -0.25 -0.20 -0.28 -0.25 -0.35 -0.11 -0.18 -0.10 -0.07
    0 68 62 WARNING: "gauden.c", line 1343: Scaling factor too small: -34801.
    931330
    ERROR: "backward.c", line 1019: alpha(1.561918e-028) <> sum of alphas * betas
    (0
    .000000e+000) in frame 371
    ERROR: "baum_welch.c", line 333: arctic_0006 ignored
    utt 0.004x 2.628e upd 0.004x 1.795e fwd 0.004x 1.218e bwd 0.000x 0.000e gau
    0.0
    04x 0.321e rsts 0.000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 6 arctic_0007 456INFO: cmn.c(175): CMN: 7.51 0.21 -0.1
    0 0.14 -0.27 -0.25 -0.32 -0.20 -0.38 -0.11 -0.17 -0.05 -0.10
    0 56 52 WARNING: "gauden.c", line 1343: Scaling factor too small: -33563.
    304111
    ERROR: "backward.c", line 1019: alpha(6.216847e-024) <> sum of alphas * betas
    (0
    .000000e+000) in frame 454
    ERROR: "baum_welch.c", line 333: arctic_0007 ignored
    utt 0.007x 1.314e upd 0.007x 0.962e fwd 0.003x 1.218e bwd 0.000x 0.000e gau
    0.0
    00x 0.000e rsts 0.000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 7 arctic_0008 405INFO: cmn.c(175): CMN: 7.30 0.29 -0.1
    0 0.14 -0.26 -0.19 -0.34 -0.16 -0.31 -0.14 -0.19 -0.07 -0.11
    0 60 56 WARNING: "gauden.c", line 1343: Scaling factor too small: -32457.
    821011
    ERROR: "backward.c", line 1019: alpha(1.159387e-025) <> sum of alphas * betas
    (0
    .000000e+000) in frame 403
    ERROR: "baum_welch.c", line 333: arctic_0008 ignored
    utt 0.004x 2.372e upd 0.004x 1.923e fwd 0.004x 1.218e bwd 0.000x 0.000e gau
    0.0
    00x 0.000e rsts 0.000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 8 arctic_0009 433INFO: cmn.c(175): CMN: 7.81 0.23 -0.1
    2 0.10 -0.32 -0.19 -0.30 -0.14 -0.28 -0.13 -0.15 -0.06 -0.10
    0 60 56 WARNING: "gauden.c", line 1343: Scaling factor too small: -35139.
    604915
    ERROR: "backward.c", line 1019: alpha(4.390695e-028) <> sum of alphas * betas
    (0
    .000000e+000) in frame 431
    ERROR: "baum_welch.c", line 333: arctic_0009 ignored
    utt 0.004x 2.308e upd 0.004x 1.859e fwd 0.004x 1.218e bwd 0.000x 0.000e gau
    0.0
    04x 0.449e rsts 0.000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 9 arctic_0010 349INFO: cmn.c(175): CMN: 7.69 0.28 -0.0
    7 0.06 -0.31 -0.22 -0.34 -0.15 -0.33 -0.16 -0.14 -0.06 -0.08
    0 44 41 WARNING: "gauden.c", line 1343: Scaling factor too small: -34503.
    746771
    ERROR: "backward.c", line 1019: alpha(4.010523e-020) <> sum of alphas * betas
    (0
    .000000e+000) in frame 347
    ERROR: "baum_welch.c", line 333: arctic_0010 ignored
    utt 0.009x 0.897e upd 0.009x 0.705e fwd 0.004x 0.897e bwd 0.000x 0.000e gau
    0.0
    00x 0.000e rsts 0.000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 10 arctic_0011 428INFO: cmn.c(175): CMN: 7.97 0.25 -0.1
    5 0.07 -0.25 -0.19 -0.33 -0.17 -0.31 -0.11 -0.15 -0.01 -0.07
    0 72 66 WARNING: "gauden.c", line 1343: Scaling factor too small: -36071.
    185083
    ERROR: "backward.c", line 1019: alpha(3.066729e-033) <> sum of alphas * betas
    (0
    .000000e+000) in frame 426
    ERROR: "baum_welch.c", line 333: arctic_0011 ignored
    utt 0.007x 1.218e upd 0.007x 0.994e fwd 0.007x 0.705e bwd 0.000x 0.000e gau
    0.0
    04x 0.321e rsts 0.000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 11 arctic_0012 424INFO: cmn.c(175): CMN: 7.42 0.32 -0.0
    1 0.07 -0.18 -0.20 -0.31 -0.18 -0.32 -0.08 -0.13 -0.05 -0.09
    0 60 56 WARNING: "gauden.c", line 1343: Scaling factor too small: -32911.
    099887
    ERROR: "backward.c", line 1019: alpha(1.333247e-024) <> sum of alphas * betas
    (0
    .000000e+000) in frame 422
    ERROR: "baum_welch.c", line 333: arctic_0012 ignored
    utt 0.007x 1.154e upd 0.004x 1.795e fwd 0.004x 1.282e bwd 0.000x 0.000e gau
    0.0
    00x 0.000e rsts 0.000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 12 arctic_0013 470INFO: cmn.c(175): CMN: 7.38 0.26 -0.1
    7 0.07 -0.29 -0.13 -0.31 -0.20 -0.36 -0.12 -0.16 -0.07 -0.08
    0 56 52 WARNING: "gauden.c", line 1343: Scaling factor too small: -32780.
    186559
    ERROR: "backward.c", line 1019: alpha(6.015397e-024) <> sum of alphas * betas
    (0
    .000000e+000) in frame 468
    ERROR: "baum_welch.c", line 333: arctic_0013 ignored
    utt 0.007x 1.154e upd 0.007x 0.962e fwd 0.007x 0.641e bwd 0.000x 0.000e gau
    0.0
    00x 0.000e rsts 0.000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 13 arctic_0014 451INFO: cmn.c(175): CMN: 8.01 0.41 -0.2
    1 0.01 -0.26 -0.23 -0.31 -0.19 -0.34 -0.05 -0.19 -0.06 -0.13
    0 76 70 WARNING: "gauden.c", line 1343: Scaling factor too small: -36413.
    387306
    ERROR: "backward.c", line 1019: alpha(2.227746e-032) <> sum of alphas * betas
    (0
    .000000e+000) in frame 449
    ERROR: "baum_welch.c", line 333: arctic_0014 ignored
    utt 0.003x 2.628e upd 0.003x 2.179e fwd 0.003x 1.603e bwd 0.000x 0.000e gau
    0.0
    00x 0.000e rsts 0.000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    overall> WIN32(N/A) 0 (-0) 0.000000e+000 0.000000e+000 0.000x 1.744e
    WARNING: "accum.c", line 618: Over 500 senones never occur in the input data.
    Th
    is is normal for context-dependent untied senone training or for adaptation,
    but
    could indicate a serious problem otherwise.
    INFO: s3mixw_io.c(232): Wrote ./mixw_counts
    INFO: s3tmat_io.c(174): Wrote ./tmat_counts
    INFO: s3gau_io.c(478): Wrote ./gauden_counts with means with vars
    INFO: main.c(1037): Counts saved to . </utt_log_lik></frame_log_lik></avg_posterior_prune></avg_states_reest></avg_states_beta></avg_states_alpha></n_state_shmm></n_frame_del></n_frame_in></id></seq>

    Thanks & Regards,
    Dhiraj

     
  • Nickolay V. Shmyrev

    The model wsj_8khz is a continuous model. You need to use

    -ts2cbfn .cont.
    

    not

    -ts2cbfn .semi.
    
     
  • Dhiraj Gupta

    Dhiraj Gupta - 2011-11-25

    Thanks, those messages are gone. I am still getting one warning at bottom.
    'WARNING: "accum.c", line 618: Over 500 senones never occur in the input data.
    This is normal for context-dependent untied senone training or for adaptation,
    but could indicate a serious problem otherwise.'

    As it mentions, this warning can be ignored for adaptation, I assume it is
    okay. I will try using the O/P model files. Thanks very much for your help!

    INFO: main.c(194): Compiled on Nov 24 2011 at 14:09:23
    INFO: cmd_ln.c(691): Parsing command line:
    bw \
    -hmmdir wsj_8kHz \
    -moddeffn wsj_8kHz\mdef.txt \
    -ts2cbfn .cont. \
    -feat 1s_c_d_dd \
    -cmn current \
    -agc none \
    -dictfn arctic20.dic \
    -ctlfn arctic20.listoffiles \
    -lsnfn arctic20.transcription \
    -accumdir .

    Current configuration:

    -2passvar no no
    -abeam 1e-100 1.000000e-100
    -accumdir .
    -agc none none
    -agcthresh 2.0 2.000000e+000
    -bbeam 1e-100 1.000000e-100
    -cb2mllrfn .1cls. .1cls.
    -cepdir
    -cepext mfc mfc
    -ceplen 13 13
    -ckptintv 0
    -cmn current current
    -cmninit 8.0 8.0
    -ctlfn arctic20.listoffiles
    -diagfull no no
    -dictfn arctic20.dic
    -example no no
    -fdictfn
    -feat 1s_c_d_dd 1s_c_d_dd
    -fullsuffixmatch no no
    -fullvar no no
    -help no no
    -hmmdir wsj_8kHz
    -latdir
    -latext
    -lda
    -ldaaccum no no
    -ldadim 0 0
    -lsnfn arctic20.transcription
    -ltsoov no no
    -lw 11.5 1.150000e+001
    -maxuttlen 0 0
    -meanfn
    -meanreest yes yes
    -mixwfn
    -mixwreest yes yes
    -mllrmat
    -mmie no no
    -mmie_type rand rand
    -moddeffn wsj_8kHz\mdef.txt
    -mwfloor 0.00001 1.000000e-005
    -npart 0
    -nskip 0
    -outphsegdir
    -outputfullpath no no
    -part 0
    -pdumpdir
    -phsegdir
    -phsegext phseg phseg
    -runlen -1 -1
    -sentdir
    -sentext sent sent
    -spthresh 0.0 0.000000e+000
    -svspec
    -timing yes yes
    -tmatfn
    -tmatreest yes yes
    -topn 4 4
    -tpfloor 0.0001 1.000000e-004
    -ts2cbfn .cont.
    -varfloor 0.00001 1.000000e-005
    -varfn
    -varnorm no no
    -varreest yes yes
    -viterbi no no

    INFO: feat.c(684): Initializing feature stream to type: '1s_c_d_dd',
    ceplen=13,
    CMN='current', VARNORM='no', AGC='none'
    INFO: cmn.c(142): mean= 12.00, mean= 0.0
    INFO: main.c(283): Reading wsj_8kHz\mdef.txt
    INFO: model_def_io.c(573): Model definition info:
    INFO: model_def_io.c(574): 110883 total models defined (54 base, 110829 tri)
    INFO: model_def_io.c(575): 443532 total states
    INFO: model_def_io.c(576): 4162 total tied states
    INFO: model_def_io.c(577): 162 total tied CI states
    INFO: model_def_io.c(578): 54 total tied transition matrices
    INFO: model_def_io.c(579): 4 max state/model
    INFO: model_def_io.c(580): 4 min state/model
    INFO: s3mixw_io.c(116): Read wsj_8kHz/mixture_weights
    INFO: s3tmat_io.c(115): Read wsj_8kHz/transition_matrices
    INFO: mod_inv.c(300): inserting tprob floor 1.000000e-004 and renormalizing
    INFO: s3gau_io.c(166): Read wsj_8kHz/means
    INFO: s3gau_io.c(166): Read wsj_8kHz/variances
    INFO: gauden.c(184): 4162 total mgau
    INFO: gauden.c(158): 1 feature streams (|0|=39 )
    INFO: gauden.c(195): 8 total densities
    INFO: gauden.c(97): min_var=1.000000e-005
    INFO: gauden.c(173): compute 4 densities/frame
    INFO: main.c(395): Will reestimate mixing weights.
    INFO: main.c(397): Will reestimate means.
    INFO: main.c(399): Will reestimate variances.
    INFO: main.c(407): Will reestimate transition matrices
    INFO: main.c(420): Reading main lexicon: arctic20.dic
    INFO: lexicon.c(218): 129246 entries added from arctic20.dic
    INFO: main.c(432): Reading filler lexicon: wsj_8kHz/noisedict
    INFO: lexicon.c(218): 27 entries added from wsj_8kHz/noisedict
    INFO: corpus.c(1078): Will process all remaining utts starting at 0
    INFO: main.c(639): Reestimation: Baum-Welch
    INFO: main.c(644): Generating profiling information consumes significant CPU
    res
    ources.
    INFO: main.c(645): If you are not interested in profiling, use -timing no
    column defns
    <seq>
    <id>
    <n_frame_in>
    <n_frame_del>
    <n_state_shmm>
    <avg_states_alpha>
    <avg_states_beta>
    <avg_states_reest>
    <avg_posterior_prune>
    <frame_log_lik>
    <utt_log_lik>
    ... timing info ...
    utt> 0 arctic_0001 405INFO: cmn.c(175): CMN: 7.57 0.30 -0.1
    9 0.06 -0.32 -0.12 -0.32 -0.20 -0.32 -0.15 -0.13 -0.09 -0.07
    0 40 16 5 5 2.293645e-103 -3.212159e+000 -1.300924e+003 utt 0.019x 1.000
    e upd 0.019x 1.000e fwd 0.008x 0.994e bwd 0.008x 1.026e gau 0.008x 0.994e rsts
    0
    .000x 0.000e rstf 0.008x 1.026e rstu 0.000x 0.000e
    utt> 1 arctic_0002 442INFO: cmn.c(175): CMN: 7.51 0.30 -0.1
    8 0.15 -0.28 -0.13 -0.28 -0.19 -0.36 -0.12 -0.14 -0.02 -0.08
    0 56 17 4 3 9.492055e-103 -2.455192e+000 -1.085195e+003 utt 0.014x 1.747
    e upd 0.014x 1.747e fwd 0.011x 1.004e bwd 0.004x 3.974e gau 0.014x 7.772e rsts
    0
    .000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 2 arctic_0003 368INFO: cmn.c(175): CMN: 7.20 0.22 -0.1
    7 0.11 -0.27 -0.09 -0.28 -0.23 -0.31 -0.12 -0.09 -0.04 -0.08
    0 44 18 4 3 1.174523e-102 -7.894920e-001 -2.905331e+002 utt 0.017x 0.994
    e upd 0.017x 0.994e fwd 0.008x 0.994e bwd 0.004x 1.026e gau 0.004x 1.987e rsts
    0
    .000x 0.000e rstf 0.004x 1.026e rstu 0.000x 0.000e
    utt> 3 arctic_0004 284INFO: cmn.c(175): CMN: 7.19 0.30 -0.1
    2 0.13 -0.20 -0.18 -0.31 -0.22 -0.31 -0.15 -0.13 -0.05 -0.06
    0 32 6 3 2 7.899203e-103 -7.920770e+000 -2.249499e+003 utt 0.005x 2.051e
    upd 0.005x 2.051e fwd 0.005x 1.026e bwd 0.000x 0.000e gau 0.005x 1.026e rsts
    0.
    000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 4 arctic_0005 340INFO: cmn.c(175): CMN: 7.19 0.34 -0.2
    1 0.09 -0.26 -0.17 -0.28 -0.19 -0.34 -0.12 -0.12 -0.04 -0.07
    0 48 12 4 4 8.137092e-103 -6.175395e+000 -2.099634e+003 utt 0.014x 0.983
    e upd 0.014x 0.983e fwd 0.005x 0.962e bwd 0.009x 0.994e gau 0.041x 1.132e rsts
    0
    .000x 0.000e rstf 0.005x 0.962e rstu 0.000x 0.000e
    utt> 5 arctic_0006 373INFO: cmn.c(175): CMN: 7.73 0.55 -0.1
    7 0.01 -0.25 -0.20 -0.28 -0.25 -0.35 -0.11 -0.18 -0.10 -0.07
    0 68 21 5 6 1.169906e-102 -5.609982e+000 -2.092523e+003 utt 0.021x 1.000
    e upd 0.021x 1.000e fwd 0.013x 1.004e bwd 0.008x 0.994e gau 0.038x 0.983e rsts
    0
    .000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 6 arctic_0007 456INFO: cmn.c(175): CMN: 7.51 0.21 -0.1
    0 0.14 -0.27 -0.25 -0.32 -0.20 -0.38 -0.11 -0.17 -0.05 -0.10
    0 56 20 6 5 1.609580e-102 -3.429510e+000 -1.563857e+003 utt 0.017x 1.205
    e upd 0.017x 1.000e fwd 0.007x 0.994e bwd 0.010x 1.004e gau 0.051x 1.085e rsts
    0
    .000x 0.000e rstf 0.003x 1.026e rstu 0.000x 0.000e
    utt> 7 arctic_0008 405INFO: cmn.c(175): CMN: 7.30 0.29 -0.1
    0 0.14 -0.26 -0.19 -0.34 -0.16 -0.31 -0.14 -0.19 -0.07 -0.11
    0 60 20 4 4 6.459098e-103 -3.381196e+000 -1.369384e+003 utt 0.015x 0.994
    e upd 0.015x 0.994e fwd 0.008x 0.994e bwd 0.008x 0.994e gau 0.031x 0.970e rsts
    0
    .000x 0.000e rstf 0.004x 1.026e rstu 0.000x 0.000e
    utt> 8 arctic_0009 433INFO: cmn.c(175): CMN: 7.81 0.23 -0.1
    2 0.10 -0.32 -0.19 -0.30 -0.14 -0.28 -0.13 -0.15 -0.06 -0.10
    0 60 20 4 4 9.473066e-103 -2.969024e+000 -1.285588e+003 utt 0.018x 1.000
    e upd 0.018x 1.000e fwd 0.007x 0.994e bwd 0.007x 0.994e gau 0.018x 0.987e rsts
    0
    .000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 9 arctic_0010 349INFO: cmn.c(175): CMN: 7.69 0.28 -0.0
    7 0.06 -0.31 -0.22 -0.34 -0.15 -0.33 -0.16 -0.14 -0.06 -0.08
    0 44 13 4 4 2.447690e-103 -4.414867e+000 -1.540789e+003 utt 0.013x 0.983
    e upd 0.013x 0.983e fwd 0.004x 0.962e bwd 0.009x 0.994e gau 0.004x 0.962e rsts
    0
    .004x 0.962e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 10 arctic_0011 428INFO: cmn.c(175): CMN: 7.97 0.25 -0.1
    5 0.07 -0.25 -0.19 -0.33 -0.17 -0.31 -0.11 -0.15 -0.01 -0.07
    0 72 19 5 5 1.689097e-102 -3.992388e+000 -1.708742e+003 utt 0.018x 1.205
    e upd 0.018x 1.000e fwd 0.007x 0.994e bwd 0.011x 1.004e gau 0.011x 1.004e rsts
    0
    .000x 0.000e rstf 0.004x 1.026e rstu 0.000x 0.000e
    utt> 11 arctic_0012 424INFO: cmn.c(175): CMN: 7.42 0.32 -0.0
    1 0.07 -0.18 -0.20 -0.31 -0.18 -0.32 -0.08 -0.13 -0.05 -0.09
    0 60 20 5 5 8.882446e-103 -5.226671e+000 -2.216109e+003 utt 0.018x 1.000
    e upd 0.018x 1.000e fwd 0.011x 1.004e bwd 0.007x 0.994e gau 0.026x 0.861e rsts
    0
    .004x 1.026e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 12 arctic_0013 470INFO: cmn.c(175): CMN: 7.38 0.26 -0.1
    7 0.07 -0.29 -0.13 -0.31 -0.20 -0.36 -0.12 -0.16 -0.07 -0.08
    0 56 16 4 4 6.322804e-103 -1.856437e-002 -8.725255e+000 utt 0.013x 0.994
    e upd 0.013x 0.994e fwd 0.007x 0.994e bwd 0.007x 0.994e gau 0.013x 0.978e rsts
    0
    .000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    utt> 13 arctic_0014 451INFO: cmn.c(175): CMN: 8.01 0.41 -0.2
    1 0.01 -0.26 -0.23 -0.31 -0.19 -0.34 -0.05 -0.19 -0.06 -0.13
    0 76 22 6 5 7.408549e-103 -5.609756e+000 -2.530000e+003 utt 0.017x 1.192
    e upd 0.017x 1.192e fwd 0.010x 0.983e bwd 0.007x 1.506e gau 0.076x 1.020e rsts
    0
    .000x 0.000e rstf 0.000x 0.000e rstu 0.000x 0.000e
    overall> WIN32(N/A) 5628 (-0) -3.792022e+000 -2.134150e+004 0.016x 1.118e
    WARNING: "accum.c", line 618: Over 500 senones never occur in the input data.
    Th
    is is normal for context-dependent untied senone training or for adaptation,
    but
    could indicate a serious problem otherwise.
    INFO: s3mixw_io.c(232): Wrote ./mixw_counts
    INFO: s3tmat_io.c(174): Wrote ./tmat_counts
    INFO: s3gau_io.c(478): Wrote ./gauden_counts with means with vars
    INFO: main.c(1037): Counts saved to . </utt_log_lik></frame_log_lik></avg_posterior_prune></avg_states_reest></avg_states_beta></avg_states_alpha></n_state_shmm></n_frame_del></n_frame_in></id></seq>

    Thanks & Regards,
    Dhiraj

     
  • Dhiraj Gupta

    Dhiraj Gupta - 2011-11-25

    Hi,

    I have a question. I got 3 outputs file tmat_counts, mixw_counts,
    gauden_counts. Below is the message.

    INFO: s3mixw_io.c(232): Wrote ./mixw_counts
    INFO: s3tmat_io.c(174): Wrote ./tmat_counts
    INFO: s3gau_io.c(478): Wrote ./gauden_counts with means with vars
    INFO: main.c(1037): Counts saved to .

    It seems that gauden_counts has both means and variances? Is that true? If
    yes, how can create 2 separate files out of it? I couldn;t find that
    information at the below link. What I understand is after I have those 2 files
    separately, I will need to run map_adapt. Is my understanding correct?

    http://cmusphinx.sourceforge.net/wiki/tutorialadapt

    Thanks & Regards,
    Dhiraj

     
  • Dhiraj Gupta

    Dhiraj Gupta - 2011-11-25

    I tried searching on this and came across the below thread. But this too
    didn't have the anser as the person who raised it said he got the answer
    himself. But I still am not able to figure out the answer :-(

    [https://sourceforge.net/projects/cmusphinx/forums/forum/382337/topic/3946031]
    (https://sourceforge.net/projects/cmusphinx/forums/forum/382337/topic/3946031)

    Thanks & REgards,
    Dhiraj

     
  • Nickolay V. Shmyrev

    It seems that gauden_counts has both means and variances?

    It contains modifications (counts) for means and variances, not means and
    variances themselves

    If yes, how can create 2 separate files out of it?

    You don't need separate files. Map_adapt will load gauden_counts from the
    current directory according to accumdir option. You only need to run a command
    from the tutorial.

    What I understand is after I have those 2 files separately, I will need to
    run map_adapt. Is my understanding correct?

    No, you are wrong.

     

Log in to post a comment.