there is a pocketsphinx plugin for gstreamer. In this case, pocketsphinx can
be launched only with the 3 main parameters (hmm, lm, dic) but not with
others. is there a new version of this gst plugin to handle all parameters???
We dont think it's needed. You can drop -mdef by renaming mdef file to
standard "mdef". You can ddro 1s_c_d_dd and you can put "-agc emax" in
standard feat.params file. So it will work out-of-box.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
in the hmm directory:
-i've copied the mdef file and rename it to 'mdef'
- i ve created a feat.params file and put the line -agc emax
it works but the result is bad.
There are some differences between the version based on
pocketsphinx_continuous and the python version with gst-pocketsphinx, for
examples:
-samprate 16000 8.000000e+03 (with gst)
-samprate 16000 1.600000e+04 (with pocketsphinx_contiuous)
Could be the explanation ??
I cannot add the parameter in the feat.params file or in the python program,
it failed!!
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
No, does not work at all.
in the gstpocketsphinx.c: "-samprate", "16000" , same result.
i ve removed other parameters in the sources to have the same default
parameters: Same result
(i check that parameters change for any modifications)
my dictionnary is an otan alphanet... when i say bravo the B appears, tango
for T, etc.... works well with the command line.
I don't know how to go ahead??
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
i have change to rate = (int) 16000) .... it does not work at all. nothing is
decoded. the recording seems not working.
i went back to the original version with english LM and DICT, i checked it
works. If i change to :
rate = (int) 16000. Nothing works!!!
Same, if i change the sample rate conversion (rate =int 16000) and the
pocketsphinx parameters ('samprate 16000' + -nfft 512???) to be coherent.
I try the example in C from this url http://www.speech.cs.cmu.edu/cmusphinx/m
oinmoin/PocketSphinxMigration
It work well with the french LM & DICT. But i don't have any example to do a
live decoding. I don't know how to do this.
The usage of gstreamer was a solution by using the vader plugin, especially in
python which is easier, unfortunately as you see it does not work.
Laurent
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
yes for pockecsphinx element and options.
not for the vader plugin.
I did not see any parameters to change the rate for vader. Maybe in the
source.
I m going to see...
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
it is better. i have a result, it is better than to have nothing :)
the rate is 16000 on all components (vader, pocketshinx plugin & options), i
have the same result as before the rate was 8000 everywhere: it is not
accurate at all.
I ve also removed some options in the pocketsphinx plugin (-nfft, -cmn,
-fwdflat, -bestpath, -maxhmmpf, -maxwpf), and i'm now quite sure pocketsphinx
is launched with same parameters in both case... i think the problem comes
from the gstreamer pipeline, not pocketsphinx, but not sure.
Regards,
Laurent
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thanks, don't want to rush you. Sure the solution in python will be welcome :)
seems very complicate to develop C to have the same thing like
python+gstreamer+vader
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
there is a pocketsphinx plugin for gstreamer. In this case, pocketsphinx can
be launched only with the 3 main parameters (hmm, lm, dic) but not with
others. is there a new version of this gst plugin to handle all parameters???
ex:
-mdef french_f0.5725.mdef -hmm french_f0 -dict frenchWords62K.dic -lm french3g62K.DMP -fdict frenchFillers.dic -feat 1s_c_d_dd -agc emax -samprate 16000
Hi
We dont think it's needed. You can drop -mdef by renaming mdef file to
standard "mdef". You can ddro 1s_c_d_dd and you can put "-agc emax" in
standard feat.params file. So it will work out-of-box.
hi,
in the hmm directory:
-i've copied the mdef file and rename it to 'mdef'
- i ve created a feat.params file and put the line -agc emax
it works but the result is bad.
There are some differences between the version based on
pocketsphinx_continuous and the python version with gst-pocketsphinx, for
examples:
-samprate 16000 8.000000e+03 (with gst)
-samprate 16000 1.600000e+04 (with pocketsphinx_contiuous)
Could be the explanation ??
I cannot add the parameter in the feat.params file or in the python program,
it failed!!
Hm,
"-samprate", "8000",
is in gstpocketsphinx.c. It seems you have to edit sample rate inside
pocketsphinx sources.
No, does not work at all.
in the gstpocketsphinx.c: "-samprate", "16000" , same result.
i ve removed other parameters in the sources to have the same default
parameters: Same result
(i check that parameters change for any modifications)
my dictionnary is an otan alphanet... when i say bravo the B appears, tango
for T, etc.... works well with the command line.
I don't know how to go ahead??
Hm, but gstreamer component also has the following:
Which causes sample rate conversion. Did you change that?
i have change to rate = (int) 16000) .... it does not work at all. nothing is
decoded. the recording seems not working.
i went back to the original version with english LM and DICT, i checked it
works. If i change to :
rate = (int) 16000. Nothing works!!!
Same, if i change the sample rate conversion (rate =int 16000) and the
pocketsphinx parameters ('samprate 16000' + -nfft 512???) to be coherent.
I try the example in C from this url http://www.speech.cs.cmu.edu/cmusphinx/m
oinmoin/PocketSphinxMigration
It work well with the french LM & DICT. But i don't have any example to do a
live decoding. I don't know how to do this.
The usage of gstreamer was a solution by using the vader plugin, especially in
python which is easier, unfortunately as you see it does not work.
Laurent
Hm, I probably need to look at this myself. From a quick look you need to put
16000 in three places:
vader element, pocketsphinx element, pocketsphinx options
Did you change vader?
yes for pockecsphinx element and options.
not for the vader plugin.
I did not see any parameters to change the rate for vader. Maybe in the
source.
I m going to see...
it is better. i have a result, it is better than to have nothing :)
the rate is 16000 on all components (vader, pocketshinx plugin & options), i
have the same result as before the rate was 8000 everywhere: it is not
accurate at all.
I ve also removed some options in the pocketsphinx plugin (-nfft, -cmn,
-fwdflat, -bestpath, -maxhmmpf, -maxwpf), and i'm now quite sure pocketsphinx
is launched with same parameters in both case... i think the problem comes
from the gstreamer pipeline, not pocketsphinx, but not sure.
Regards,
Laurent
I give up the usage of python for the moment.
I tested successfully pocketsphinx in C, but i did not find any examples to do
a live decoding.
My goal is still to have a LIVE french recognition with a small dictionary...
could you help me?
Laurent
Heh, I need to find some time to check this
About C example, take a look at this thread:
https://sourceforge.net/projects/cmusphinx/forums/forum/5471/topic/3514523
Thanks, don't want to rush you. Sure the solution in python will be welcome :)
seems very complicate to develop C to have the same thing like
python+gstreamer+vader
I am sorry. I posted it again in the correct forum.