Acoustic Model Discussions

Flat
Using HTK acoustic Model on pocketsphinx
User: krlospatrick
Date: 9/27/2010 2:46 pm
Views: 7830
Rating: 4

Hi All,

I have been used HTK for many years, and now I'm trying to test pocketsphinx decoder. I have built a 16 gaussians mixture MFCC_E_D_A_Z cross-word triphones acoustic model using 15 hours of data in HTK. I had great results with HDecode.

I used the Python HTK to Sphinx acoustic model converter:

http://cmusphinx.sourceforge.net/2010/08/python-htk-converter/

It' seems to work! But when I go to test it on pocketsphinx I got the follow error message:

--------------------------------------------------------------------

pocketsphinx_batch \
        -adcin yes \
        -hmm LapsAM-1.5-16k-Sphinx \
        -lw 10 \
        -feat 1s_c_d_dd \
        -dict Sphinx/dictionary.dic \
        -lm /sphinx/test_benchmark/etc/laps1.5.lm \
        -ctl wav-list.txt \
        -cepdir sphinx/test_benchmark/feat \
        -cepext .wav \
        -hyp /sphinx/test_benchmark/result/laps1.5-1-1.match \
        -agc none \
        -varnorm no \
        -cmn current \
        -ceplen 13 \
        -wlen 0.025

Current configuration:
[NAME]          [DEFLT]         [VALUE]
-adchdr         0               0
-adcin          no              yes
-agc            none            none
-agcthresh      2.0             2.000000e+00
-alpha          0.97            9.700000e-01
-argfile
-ascale         20.0            2.000000e+01
-backtrace      no              no
-beam           1e-48           1.000000e-48
-bestpath       yes             yes
-bestpathlw     9.5             9.500000e+00
-bghist         no              no
-build_outdirs  yes             yes
-cepdir                         /home/04080002801/sphinx/test_benchmark/feat
-cepext         .mfc            .wav
-ceplen         13              13
-cmn            current         current
-cmninit        8.0             8.0
-compallsen     no              no
-ctl                            wav-list.txt
-ctlcount       -1              -1
-ctlincr        1               1
-ctloffset      0               0
-ctm
-debug                          0
-dict                           /home/04080002801/htkEmbedded/Sphinx/dictionary.dic
-dictcase       no              no
-dither         no              no
-doublebw       no              no
-ds             1               1
-fdict
-feat           1s_c_d_dd       1s_c_d_dd
-featparams
-fillprob       1e-8            1.000000e-08
-frate          100             100
-fsg
-fsgctl
-fsgdir
-fsgext
-fsgusealtpron  yes             yes
-fsgusefiller   yes             yes
-fwdflat        yes             yes
-fwdflatbeam    1e-64           1.000000e-64
-fwdflatefwid   4               4
-fwdflatlw      8.5             8.500000e+00
-fwdflatsfwin   25              25
-fwdflatwbeam   7e-29           7.000000e-29
-fwdtree        yes             yes
-hmm                            /home/04080002801/LapsAM-1.5-16k-Sphinx
-hyp                            /home/04080002801/sphinx/test_benchmark/result/laps1.5-1-1.match
-hypseg
-input_endian   little          little
-jsgf
-kdmaxbbi       -1              -1
-kdmaxdepth     0               0
-kdtree
-latsize        5000            5000
-lda
-ldadim         0               0
-lextreedump    0               0
-lifter         0               0
-lm                             /home/04080002801/sphinx/test_benchmark/etc/laps1.5.lm
-lmctl
-lmname         default         default
-lmnamectl
-logbase        1.0001          1.000100e+00
-logfn
-logspec        no              no
-lowerf         133.33334       1.333333e+02
-lpbeam         1e-40           1.000000e-40
-lponlybeam     7e-29           7.000000e-29
-lw             6.5             1.000000e+01
-maxhmmpf       -1              -1
-maxnewoov      20              20
-maxwpf         -1              -1
-mdef
-mean
-mfclogdir
-mixw
-mixwfloor      0.0000001       1.000000e-07
-mllr
-mllrctl
-mllrdir
-mllrext
-mmap           yes             yes
-nbest          0               0
-nbestdir
-nbestext       .hyp            .hyp
-ncep           13              13
-nfft           512             512
-nfilt          40              40
-nwpen          1.0             1.000000e+00
-outlatdir
-pbeam          1e-48           1.000000e-48
-pip            1.0             1.000000e+00
-pl_beam        1e-10           1.000000e-10
-pl_pbeam       1e-5            1.000000e-05
-pl_window      0               0
-rawlogdir
-remove_dc      no              no
-round_filters  yes             yes
-samprate       16000           1.600000e+04
-seed           -1              -1
-sendump
-senmgau
-silprob        0.005           5.000000e-03
-smoothspec     no              no
-svspec
-tmat
-tmatfloor      0.0001          1.000000e-04
-topn           4               4
-topn_beam      0               0
-toprule
-transform      legacy          legacy
-unit_area      yes             yes
-upperf         6855.4976       6.855498e+03
-usewdphones    no              no
-uw             1.0             1.000000e+00
-var
-varfloor       0.0001          1.000000e-04
-varnorm        no              no
-verbose        no              no
-warp_params
-warp_type      inverse_linear  inverse_linear
-wbeam          7e-29           7.000000e-29
-wip            0.65            6.500000e-01
-wlen           0.025625        2.500000e-02

INFO: feat.c(848): Initializing feature stream to type: '1s_c_d_dd', ceplen=13, CMN='current', VARNORM='no', AGC='none'
INFO: cmn.c(142): mean[0]= 12.00, mean[1..12]= 0.0
INFO: mdef.c(520): Reading model definition: /home/04080002801/LapsAM-1.5-16k-Sphinx/mdef
FATAL_ERROR: "mdef.c", line 299: CI-senone-id(337) > #CI-senones(117): aa       -       -       -       n/a     55      337     338     112     N

Someone tried to use this converter? works ?

thanks for any help!

[ ]'s

 

Patrick

 

--- (Edited on 9/27/2010 2:46 pm [GMT-0500] by krlospatrick) ---

Re: Using HTK acoustic Model on pocketsphinx
User: nsh
Date: 9/28/2010 8:10 am
Views: 108
Rating: 4

Hello Patrick

Please take into account that this code is hightly exerimental. It does work in some case, but there are several issues known

1. Accuracy after conversion is a little bit worse than with HDecode (2% relatively worse). We are looking on this issue now.

2. Feature extraction is not compatible. You need to use other feature extarction options to be the same as HTK. You used incorrect ones in your command line. You need to add -dct htk -nfilter 22 and set up upperf and lowerf to something like 1 and 8000. Alternatively you can process HTK mfcc file.

3. Your problem reveals that there is a bug in the convertor that fails to convert the model properly. I need to be able to reproduce your issue in order to fix it in htk2s3conv sources. Specifically I need the model files you have. This issue is tracked in CMUSphinx bug tracking system:

https://sourceforge.net/tracker/?func=detail&aid=3077225&group_id=1904&atid=101904

It might be easier for you to retrain your model using SphinxTrain, i think you'll enjoy the results.I also respectively recommend you to ask further questions on CMUSphinx forums, the default support place for CMUSphinx project.

--- (Edited on 9/28/2010 17:10 [GMT+0400] by nsh) ---

Re: Using HTK acoustic Model on pocketsphinx
User: Patrick
Date: 9/28/2010 1:21 pm
Views: 3910
Rating: 1

Hi Nickolay,

Thanks for your answer!

I follow the CMUSphinx bug tracking system:

https://sourceforge.net/tracker/?func=detail&aid=3077225&group_id=1904&atid=101904

>> I need to be able to reproduce your issue in order to fix it in htk2s3conv sources. Specifically I need the model files you have.

This is the link for the original HTK acoustic Model:

http://www.laps.ufpa.br/patrick/downloads/LapsAM-1.5-16k-Pre-Sphinx.rar

You can use it to test the conversion scripts. Please, let me know any improvement. I'll try to train the model with the SphinxTrain.

Thanks for the help !!

Patrick

 

--- (Edited on 9/28/2010 1:21 pm [GMT-0500] by Visitor) ---

PreviousNext