compiled pocketsphinx.js does not work about pocketsphinx.js

[syl22-00]

@AakashGfude , I do get the same warnings with recent versions of emscripten, but they do not seem to have any consequence as the generated JavaScript file works fine. Do you see any error in the JavaScript console? To be able to help you more, could you:

• Send the output you get from the JavaScript console?
• Send the exact set of commands you run to generate pocketsphinx.js.
• Share your generated pocketsphinx.js file somewhere.

[AakashGfude]

To generate pocketsphinx.js , the commands i used were:

cmake -DEMSCRIPTEN=1 -DCMAKE_TOOLCHAIN_FILE=/home/aakash2/emscripten/cmake/Platform/Emscripten.cmake -DHMM_BASE=/home/aakash2/HMM_BASE -DHMM_FOLDERS=model1 ..
output was ::
HMM_BASE is /home/aakash2/HMM_BASE
HMM_FOLDERS is model1
Copying /home/aakash2/HMM_BASE/model1 to binary dir
-- Configuring done
-- Generating done
-- Build files have been written to: /home/aakash2/build
then used 'make' and the warnings showed up.

and my pocketsphinx.js file is 125 MB , which is too much..
the link for pocketsphinx.js is : https://drive.google.com/file/d/0B6Pc1k2W4-laWVUwT3pITmhIb0k/edit?usp=sharing

Thanks a lot,

[syl22-00]

Yes, your file is too large, the browsers won't load it:

• Firefox says FUNCTION_TABLE[HEAP32[((HEAP32[(r25 >> 2)] + 16) >> 2)]] is not a function
• Chrome says Uncaught Error: Uncaught RangeError: Maximum call stack size exceeded

It seems like your acoustic model is very large, can you make it smaller, with less senones? I stayed with 200 senones in the default packaged model.
Another thing I noticed is that you have both sendump and mixture_weigths. You don't need both, they are the same (one is binary, one is text), you'd rather keep sendump. Half of your file is taken by mixture_weights.

I hope that helps.

[AakashGfude]

Thanks,
I removed mixture_weights and mdef.txt and now it is 31.4 MB , still very large though . Sorry i am very new to speech recognition, it has been 2 weeks when i started , so, can u give me any clue as to how to reduce the senones , i had added only recordings of 25 small sentences to the default model " hub4wsj_sc_8k "
And once we are done compiling , we just have to replace your pocketsphinx.js in the webapp directory with ours r8 ?

[syl22-00]

Yes, @AakashGfude , you can just remove mixture_weights if you have sendump, but you definitely need mdef (not sure what you meant with mdef.txt).

hub4 is built on 8kHz audio, so I'd recommend you to build, for instance, a RM1 model as explained in the cmusphinx model training tutorial, which uses 16kHz audio. You can then go through adaptation if you have files to use for that. Note that with adaptation, you just add a small file to your acoustic model, so you can't change your number of senones during adaptation. You can also just start adaptation from the RM1 model provided with pocketsphinx.js.

And yes, you can then just use it in the demo by replacing pocketsphinx.js, but:

• As explained in the docs, if your model is built on 8kHz audio, you need to provide 8kHz audio with the recorder (see recorder doc).
• The phone set you use must be the same as the one in the demo (which should be the case if you use hub4 or RM1, all based on the CMU dictionary).

[AakashGfude]

Theres an mdef.txt which comes which hub4 which we have to convert to binary just like mixture_weights and sendump according to that tutorial .

Thanks a lot mate .

[AakashGfude]

One last issue..
I had now compiled the default hub4 model without any adaptation and the pocketsphinx.js file is still 31.4 MB. It has much greater senones i guess. But when i replace it with urs in the webapp folder and put that in my local web server and open the demo webpage, the recognizer doesnt get ready.. Can u plz just check it once.. i changed the outputsamplerate in audiorecorder and audiorecorderworker to 8000 but still it doesnt work .

I just wanted to know if size is causing the problem, means i should not use hub4 or i have done something wrong in the compilation process..
If size is the problem then i will go to rm1 model . else i have to look into the building process..

Thanks,

[syl22-00]

@AakashGfude can you please post the output from the JavaScript console? It's not possible to guess what happens otherwise.

Meanwhile, I tried this:

$ mkdir build
$ cd build
$ cmake -DEMSCRIPTEN=1 -DCMAKE_TOOLCHAIN_FILE=/path/to/emscripten/cmake/Platform/Emscripten_unix.cmake -DHMM_BASE=/path/to/cmusphinx/pocketsphinx/model/hmm/en_US/ -DHMM_FOLDERS=hub4wsj_sc_8k ..
$ make

Which gave me a 22MB JavaScript file, replaced the one in webapp/js.

In webapp/live.html, I added outputSampleRate: 8000 in audioRecorderConfig.

Then I opened live.html and everything went fine.

[AakashGfude]

The output from the javascript console is ::

Current configuration: js/pocketsphinx.js:150
[NAME] [DEFLT] [VALUE] js/pocketsphinx.js:150
-adcdev js/pocketsphinx.js:150
-agc none none js/pocketsphinx.js:150
-agcthresh 2.0 2.000000e+00 js/pocketsphinx.js:150
-alpha 0.97 9.700000e-01 js/pocketsphinx.js:150
-argfile js/pocketsphinx.js:150
-ascale 20.0 2.000000e+01 js/pocketsphinx.js:150
-aw 1 1 js/pocketsphinx.js:150
-backtrace no no js/pocketsphinx.js:150
-beam 1e-48 1.000000e-48 js/pocketsphinx.js:150
-bestpath yes no js/pocketsphinx.js:150
-bestpathlw 9.5 9.500000e+00 js/pocketsphinx.js:150
-bghist no no js/pocketsphinx.js:150
-ceplen 13 13 js/pocketsphinx.js:150
-cmn current current js/pocketsphinx.js:150
-cmninit 8.0 8.0 js/pocketsphinx.js:150
-compallsen no no js/pocketsphinx.js:150
-debug 0 js/pocketsphinx.js:150
-dict js/pocketsphinx.js:150
-dictcase no no js/pocketsphinx.js:150
-dither no no js/pocketsphinx.js:150
-doublebw no no js/pocketsphinx.js:150
-ds 1 1 js/pocketsphinx.js:150
-fdict js/pocketsphinx.js:150
-feat 1s_c_d_dd 1s_c_d_dd js/pocketsphinx.js:150
-featparams js/pocketsphinx.js:150
-fillprob 1e-8 1.000000e-08 js/pocketsphinx.js:150
-frate 100 100 js/pocketsphinx.js:150
-fsg js/pocketsphinx.js:150
-fsgusealtpron yes yes js/pocketsphinx.js:150
-fsgusefiller yes yes js/pocketsphinx.js:150
-fwdflat yes yes js/pocketsphinx.js:150
-fwdflatbeam 1e-64 1.000000e-64 js/pocketsphinx.js:150
-fwdflatefwid 4 4 js/pocketsphinx.js:150
-fwdflatlw 8.5 8.500000e+00 js/pocketsphinx.js:150
-fwdflatsfwin 25 25 js/pocketsphinx.js:150
-fwdflatwbeam 7e-29 7.000000e-29 js/pocketsphinx.js:150
-fwdtree yes yes js/pocketsphinx.js:150
-hmm model js/pocketsphinx.js:150
-infile js/pocketsphinx.js:150
-input_endian little little js/pocketsphinx.js:150
-jsgf js/pocketsphinx.js:150
-kdmaxbbi -1 -1 js/pocketsphinx.js:150
-kdmaxdepth 0 0 js/pocketsphinx.js:150
-kdtree js/pocketsphinx.js:150
-latsize 5000 5000 js/pocketsphinx.js:150
-lda js/pocketsphinx.js:150
-ldadim 0 0 js/pocketsphinx.js:150
-lextreedump 0 0 js/pocketsphinx.js:150
-lifter 0 0 js/pocketsphinx.js:150
-lm js/pocketsphinx.js:150
-lmctl js/pocketsphinx.js:150
-lmname default default js/pocketsphinx.js:150
-logbase 1.0001 1.000100e+00 js/pocketsphinx.js:150
-logfn js/pocketsphinx.js:150
-logspec no no js/pocketsphinx.js:150
-lowerf 133.33334 1.333333e+02 js/pocketsphinx.js:150
-lpbeam 1e-40 1.000000e-40 js/pocketsphinx.js:150
-lponlybeam 7e-29 7.000000e-29 js/pocketsphinx.js:150
-lw 6.5 6.500000e+00 js/pocketsphinx.js:150
-maxhmmpf -1 -1 js/pocketsphinx.js:150
-maxnewoov 20 20 js/pocketsphinx.js:150
-maxwpf -1 -1 js/pocketsphinx.js:150
-mdef js/pocketsphinx.js:150
-mean js/pocketsphinx.js:150
-mfclogdir js/pocketsphinx.js:150
-min_endfr 0 0 js/pocketsphinx.js:150
-mixw js/pocketsphinx.js:150
-mixwfloor 0.0000001 1.000000e-07 js/pocketsphinx.js:150
-mllr js/pocketsphinx.js:150
-mmap yes yes js/pocketsphinx.js:150
-ncep 13 13 js/pocketsphinx.js:150
-nfft 512 512 js/pocketsphinx.js:150
-nfilt 40 40 js/pocketsphinx.js:150
-nwpen 1.0 1.000000e+00 js/pocketsphinx.js:150
-pbeam 1e-48 1.000000e-48 js/pocketsphinx.js:150
-pip 1.0 1.000000e+00 js/pocketsphinx.js:150
-pl_beam 1e-10 1.000000e-10 js/pocketsphinx.js:150
-pl_pbeam 1e-5 1.000000e-05 js/pocketsphinx.js:150
-pl_window 0 0 js/pocketsphinx.js:150
-rawlogdir js/pocketsphinx.js:150
-remove_dc no no js/pocketsphinx.js:150
-round_filters yes yes js/pocketsphinx.js:150
-samprate 16000 1.600000e+04 js/pocketsphinx.js:150
-seed -1 -1 js/pocketsphinx.js:150
-sendump js/pocketsphinx.js:150
-senlogdir js/pocketsphinx.js:150
-senmgau js/pocketsphinx.js:150
-silprob 0.005 5.000000e-03 js/pocketsphinx.js:150
-smoothspec no no js/pocketsphinx.js:150
-svspec js/pocketsphinx.js:150
-time no no js/pocketsphinx.js:150
-tmat js/pocketsphinx.js:150
-tmatfloor 0.0001 1.000000e-04 js/pocketsphinx.js:150
-topn 4 4 js/pocketsphinx.js:150
-topn_beam 0 0 js/pocketsphinx.js:150
-toprule js/pocketsphinx.js:150
-transform legacy legacy js/pocketsphinx.js:150
-unit_area yes yes js/pocketsphinx.js:150
-upperf 6855.4976 6.855498e+03 js/pocketsphinx.js:150
-usewdphones no no js/pocketsphinx.js:150
-uw 1.0 1.000000e+00 js/pocketsphinx.js:150
-var js/pocketsphinx.js:150
-varfloor 0.0001 1.000000e-04 js/pocketsphinx.js:150
-varnorm no no js/pocketsphinx.js:150
-verbose no no js/pocketsphinx.js:150
-warp_params js/pocketsphinx.js:150
-warp_type inverse_linear inverse_linear js/pocketsphinx.js:150
-wbeam 7e-29 7.000000e-29 js/pocketsphinx.js:150
-wip 0.65 6.500000e-01 js/pocketsphinx.js:150
-wlen 0.025625 2.562500e-02 js/pocketsphinx.js:150
js/pocketsphinx.js:150
INFO: cmd_ln.c(691): Parsing command line: js/pocketsphinx.js:150
\ js/pocketsphinx.js:150
-dither yes \ js/pocketsphinx.js:150
-nfilt 20 \ js/pocketsphinx.js:150
-lowerf 1 \ js/pocketsphinx.js:150
-upperf 4000 \ js/pocketsphinx.js:150
-wlen 0.025 \ js/pocketsphinx.js:150
-transform dct \ js/pocketsphinx.js:150
-round_filters no \ js/pocketsphinx.js:150
-remove_dc yes \ js/pocketsphinx.js:150
-svspec 0-12/13-25/26-38 \ js/pocketsphinx.js:150
-feat 1s_c_d_dd \ js/pocketsphinx.js:150
-agc none \ js/pocketsphinx.js:150
-cmn current \ js/pocketsphinx.js:150
-varnorm no js/pocketsphinx.js:150
js/pocketsphinx.js:150
Current configuration: js/pocketsphinx.js:150
[NAME] [DEFLT] [VALUE] js/pocketsphinx.js:150
-agc none none js/pocketsphinx.js:150
-agcthresh 2.0 2.000000e+00 js/pocketsphinx.js:150
-alpha 0.97 9.700000e-01 js/pocketsphinx.js:150
-ceplen 13 13 js/pocketsphinx.js:150
-cmn current current js/pocketsphinx.js:150
-cmninit 8.0 8.0 js/pocketsphinx.js:150
-dither no yes js/pocketsphinx.js:150
-doublebw no no js/pocketsphinx.js:150
-feat 1s_c_d_dd 1s_c_d_dd js/pocketsphinx.js:150
-frate 100 100 js/pocketsphinx.js:150
-input_endian little little js/pocketsphinx.js:150
-lda js/pocketsphinx.js:150
-ldadim 0 0 js/pocketsphinx.js:150
-lifter 0 0 js/pocketsphinx.js:150
-logspec no no js/pocketsphinx.js:150
-lowerf 133.33334 1.000000e+00 js/pocketsphinx.js:150
-ncep 13 13 js/pocketsphinx.js:150
-nfft 512 512 js/pocketsphinx.js:150
-nfilt 40 20 js/pocketsphinx.js:150
-remove_dc no yes js/pocketsphinx.js:150
-round_filters yes no js/pocketsphinx.js:150
-samprate 16000 1.600000e+04 js/pocketsphinx.js:150
-seed -1 -1 js/pocketsphinx.js:150
-smoothspec no no js/pocketsphinx.js:150
-svspec 0-12/13-25/26-38 js/pocketsphinx.js:150
-transform legacy dct js/pocketsphinx.js:150
-unit_area yes yes js/pocketsphinx.js:150
-upperf 6855.4976 4.000000e+03 js/pocketsphinx.js:150
-varnorm no no js/pocketsphinx.js:150
-verbose no no js/pocketsphinx.js:150
-warp_params js/pocketsphinx.js:150
-warp_type inverse_linear inverse_linear js/pocketsphinx.js:150
-wlen 0.025625 2.500000e-02 js/pocketsphinx.js:150
js/pocketsphinx.js:150
INFO: acmod.c(246): Parsed model-specific feature parameters from model/feat.params js/pocketsphinx.js:150
INFO: fe_interface.c(299): You are using the internal mechanism to generate the seed. js/pocketsphinx.js:150
INFO: feat.c(713): Initializing feature stream to type: '1s_c_d_dd', ceplen=13, CMN='current', VARNORM='no', AGC='none' js/pocketsphinx.js:150
INFO: cmn.c(142): mean[0]= 12.00, mean[1..12]= 0.0 js/pocketsphinx.js:150
INFO: acmod.c(167): Using subvector specification 0-12/13-25/26-38 js/pocketsphinx.js:150
INFO: mdef.c(517): Reading model definition: model/mdef js/pocketsphinx.js:150
INFO: mdef.c(528): Found byte-order mark BMDF, assuming this is a binary mdef file js/pocketsphinx.js:150
INFO: bin_mdef.c(336): Reading binary model definition: model/mdef js/pocketsphinx.js:150
INFO: bin_mdef.c(513): 50 CI-phone, 143047 CD-phone, 3 emitstate/phone, 150 CI-sen, 5150 Sen, 27135 Sen-Seq js/pocketsphinx.js:150
INFO: tmat.c(205): Reading HMM transition probability matrices: model/transition_matrices js/pocketsphinx.js:150
INFO: acmod.c(121): Attempting to use SCHMM computation module js/pocketsphinx.js:150
INFO: ms_gauden.c(198): Reading mixture gaussian parameter: model/means js/pocketsphinx.js:150
INFO: ms_gauden.c(292): 1 codebook, 3 feature, size: js/pocketsphinx.js:150
3
INFO: ms_gauden.c(294): 256x13 js/pocketsphinx.js:150
INFO: ms_gauden.c(198): Reading mixture gaussian parameter: model/variances js/pocketsphinx.js:150
INFO: ms_gauden.c(292): 1 codebook, 3 feature, size: js/pocketsphinx.js:150
3
INFO: ms_gauden.c(294): 256x13 js/pocketsphinx.js:150
INFO: ms_gauden.c(354): 0 variance values floored js/pocketsphinx.js:150
INFO: s2_semi_mgau.c(903): Loading senones from dump file model/sendump js/pocketsphinx.js:150
INFO: s2_semi_mgau.c(927): BEGIN FILE FORMAT DESCRIPTION js/pocketsphinx.js:150
INFO: s2_semi_mgau.c(990): Rows: 256, Columns: 5150 js/pocketsphinx.js:150
INFO: s2_semi_mgau.c(1022): Using memory-mapped I/O for senones js/pocketsphinx.js:150
INFO: s2_semi_mgau.c(1296): Maximum top-N: 4 Top-N beams: 0 0 0 js/pocketsphinx.js:150
INFO: dict.c(317): Allocating 4107 * 20 bytes (80 KiB) for word entries js/pocketsphinx.js:150
INFO: dict.c(341): Reading filler dictionary: model/noisedict js/pocketsphinx.js:150
INFO: dict.c(211): Allocated 0 KiB for strings, 0 KiB for phones js/pocketsphinx.js:150
INFO: dict.c(344): 11 words read js/pocketsphinx.js:150
INFO: dict2pid.c(396): Building PID tables for dictionary js/pocketsphinx.js:150
INFO: dict2pid.c(404): Allocating 50^3 * 2 bytes (244 KiB) for word-initial triphones js/pocketsphinx.js:150
INFO: dict2pid.c(131): Allocated 30200 bytes (29 KiB) for word-final triphones js/pocketsphinx.js:150
INFO: dict2pid.c(195): Allocated 30200 bytes (29 KiB) for single-phone word triphones js/pocketsphinx.js:150
Uncaught TypeError: undefined is not a function

U have any idea what is causing that error.

[AakashGfude]

I used your mixture_weight. file which you had removed .

Here is my compiled javascript file of rm1_200 .. if you want to have a look :

https://drive.google.com/folderview?id=0B6Pc1k2W4-laWTJGNVdFTnZsejQ&usp=sharing

same error..

[hanksmathew]

Hi , I had a similar kind of error . I copied your CMakeLists.txt and put it in the directory where i had pocketsphinx and sphinxbase . when running cmake it showed fe_noise.c file is missing and my sphinxbase didnt have it so i removed it from the sphinxbase source files in the CMakeLists.txt file .
And i am getting similar behaviour from pocketsphinx.js file .

[syl22-00]

@hanksmathew pocketsphinx.js includes the sources of sphinxbase and pocketsphinx, you should use those source files, they are up to date and in sync with CMakeLists.txt.

You were probably working with an old version of pocketsphinx/sphinxbase as fe_noise.c was added fairly recently.

[syl22-00]

@AakashGfude I looked at your file. The recognizer actually loads successfully, but the call to addWords fails, so the acoustic model is probably bad.

I do not know what model files you packaged, I compared your values with mine and 3 of the 7 files of the model were different. It is a file that you have trained yourself on RM1?

[AakashGfude]

yess.. i was doing adaptation with recordings of these 26 sentences https://drive.google.com/file/d/0B6Pc1k2W4-laNmNxd0c4Z095SVE/edit?usp=sharing
and created .dic and .lm file with lmtool . Created the .transcription and .fileids according to the format given in http://cmusphinx.sourceforge.net/wiki/tutorialadapt and followed all the steps till the creation of sendump according to the link .

And yaa i copied CMakeLists.txt and the fe folder in your libsphinxbase directory of sphinxbase to my compiled sphinxbase which i downloaded from the downloads menu of that site , as it didnt have fe_noise.c as @hanksmathew said.. maybe cmusphinx.sourceforge.net has an old sphinxbase.. and i ran cmake in a build folder , then ran make.

Did i do anything wrong ?

[AakashGfude]

Now, i downloaded a different sphinbase and pocketsphinx and the make check commnad says that they have passed all the tests .. and this time sphinxbase has all the latest files like fe_noise and fe_type.h .
but after adapting acoustic model for hub4, generating makefile, when i run make on the build folder, i get the following warnings and error :

In file included from /home/aakash2/sphinxbase/src/libsphinxbase/fe/fe_noise.c:60:
/home/aakash2/sphinxbase/src/libsphinxbase/fe/fe_internal.h:68:17: warning:
redefinition of typedef 'frame_t' is a C11 feature
[-Wtypedef-redefinition]
typedef float64 frame_t;
^
/home/aakash2/sphinxbase/src/libsphinxbase/fe/fe_type.h:59:17: note: previous
definition is here
typedef float64 frame_t;
^
In file included from /home/aakash2/sphinxbase/src/libsphinxbase/fe/fe_noise.c:60:
/home/aakash2/sphinxbase/src/libsphinxbase/fe/fe_internal.h:69:17: warning:
redefinition of typedef 'powspec_t' is a C11 feature
[-Wtypedef-redefinition]
typedef float64 powspec_t;
^
/home/aakash2/sphinxbase/src/libsphinxbase/fe/fe_type.h:60:17: note: previous
definition is here
typedef float64 powspec_t;
^
In file included from /home/aakash2/sphinxbase/src/libsphinxbase/fe/fe_noise.c:60:
/home/aakash2/sphinxbase/src/libsphinxbase/fe/fe_internal.h:70:17: warning:
redefinition of typedef 'window_t' is a C11 feature
[-Wtypedef-redefinition]
typedef float64 window_t;
^
/home/aakash2/sphinxbase/src/libsphinxbase/fe/fe_type.h:61:17: note: previous
definition is here
typedef float64 window_t;
^
In file included from /home/aakash2/sphinxbase/src/libsphinxbase/fe/fe_noise.c:60:
/home/aakash2/sphinxbase/src/libsphinxbase/fe/fe_internal.h:71:34: error:
typedef redefinition with different types ('struct complex' vs
'struct complex')
typedef struct { float64 r, i; } complex;
^
/home/aakash2/sphinxbase/src/libsphinxbase/fe/fe_type.h:62:34: note: previous
definition is here
typedef struct { float64 r, i; } complex;
^
3 warnings and 1 error generated.
ERROR root: compiler frontend failed to generate LLVM bitcode, halting
make[2]: *** [CMakeFiles/pocketsphinx.dir/sphinxbase/src/libsphinxbase/fe/fe_noise.c.o] Error 1
make[1]: *** [CMakeFiles/pocketsphinx.dir/all] Error 2
make: *** [all] Error 2

If u know why this error comes then please help .

and can u share your mixture_weights file of rm1 model .

[syl22-00]

@AakashGfude , as I said earlier, the sources of sphinxbase and pocketsphinx are included in pocketsphinx.js, use these ones, do not copy your CMakeLists.txt somewhere else.

To diagnose your issue, I suggest you to:

1. Compile pocketsphinx.js with the default provided model, following the docs, and test it (run the test suite, all tests should pass).
2. Build your own acoustic model with SphinxTrain, and make sure it works, for instance with pocketsphinx_batch (compiled natively, nothing to do with pocketsphinx.js at this point)
3. Once you're sure your acoustic model is correct, compile pocketsphinx.js again with HMM_BASE and HMM_FOLDERS pointing to your new model, and test it with the test suite.
4. Try to do your adaptation. If you have issues in this process, I suggest you to ask the cmusphinx community for instance on #cmusphinx on IRC.
5. Test your adaptation data again with pocketsphinx_batch
6. Once you're sure your adapted acoustic model is correct, compile pocketsphinx.js again with HMM_BASE and HMM_FOLDERS pointing to your new model.

If you still have issues, please tell which step fails, and provide your acoustic model files and the compilation and JavaScript console outputs.

[AakashGfude]

Its working fine now :)
I was doing a lot of useless stuff .

Thanks a lottt dude .